Rethinking Longitudinal Research on Canadian Immigrant Health: Methodological Insights, Emerging Challenges, and Future Considerations

Abstract

Longitudinal research provides critical insights into the evolving health trajectories of immigrants, capturing changes from initial arrival through to long-term settlement. However, longitudinal studies on immigrant health in Canada face persistent methodological challenges that limit their impact and policy relevance. This review critically examines 34 peer-reviewed articles, published between 1996 and 2024, that employed longitudinal data to investigate physical and mental health outcomes among Canadian immigrants. We identify key methodological limitations, including a heavy reliance on earlier datasets (71% of studies used data collected between 1994 and 2007), oversimplified outcome measures (e.g., collapsing continuous or Likert-scale variables into dichotomous categories without clear justification), the limited use of appropriate longitudinal methods, and the inadequate handling of missing data. Advancing immigrant health research in Canada will require enhanced data infrastructure, greater methodological rigor, and more transparent reporting practices to better inform evidence-based policy. This review offers researchers and policymakers a clear summary of existing methodological gaps and presents practical strategies to strengthen future longitudinal research on immigrant health in Canada.

1. Introduction

Immigration has historically been a fundamental aspect of nation-building in Canada, serving as a crucial driver of both population growth and economic development. Since Confederation, immigration has played a pivotal role in shaping the country’s demographic landscape, with its importance only increasing over time (Reitz 2005; Statistics Canada 2022, 2024). Projections from Statistics Canada estimate that immigrants will comprise approximately 30% to 34% of Canada’s total population by 2041 (Statistics Canada 2022). Although the recently announced 2025–2027 Immigration Levels Plan aims to place relatively less emphasis on increasing immigration rates in the coming years, immigration still remains the primary driver of Canada’s population growth (Statistics Canada 2024). With such a continuously growing immigrant population, understanding shifts in immigrants’ post-migration health is critical, as these changes are closely linked to a potential increasing burden on Canada’s healthcare system, especially within the context of Canada’s universal healthcare system (Chronic Disease Prevention Alliance of Canada 2017). Furthermore, studies have also indicated that declining health and well-being are significant factors influencing some immigrants’ decisions to leave the host country, highlighting the urgent need to address these issues to improve immigrant retention and integration (Sheftel 2023; Handlos et al. 2015). These findings are consistent across various immigrant-receiving countries, including Canada and the United States.

A key concept in understanding immigrant health in Canada is the Healthy Immigrant Effect (HIE), which refers to the phenomenon where immigrants typically arrive in the host country in better health than the native-born population (Kennedy et al. 2015; Vang et al. 2017). However, this initial health advantage often diminishes over time, with immigrants’ health eventually converging with, or even declining below, that of the native-born population, particularly concerning chronic conditions (Newbold 2006; De Maio and Kemp 2010; Antecol and Bedard 2006). In this context, longitudinal studies are especially valuable, as they enable researchers to examine immigrants’ health and related lifestyle changes over time, providing deeper insights into the mechanisms behind this phenomenon. Although longitudinal data present certain challenges, such as high costs and participant attrition, they are invaluable for tracking individuals’ health changes over extended periods. This approach enhances the understanding of post-migration health trajectories and supports the application of theoretical frameworks like the life course perspective.

A substantial body of literature exists on immigrant health and well-being. However, most studies rely on cross-sectional data, offering only a snapshot of immigrants’ health at specific points in time. Despite the critical importance of longitudinal approaches, there remains a notable gap in research utilizing such methods to explore the HIE in Canada. This review aims to critically assess the application of quantitative longitudinal methods in immigrant health studies, identify methodological gaps, and offer recommendations to advance future research in this field. Our review is organized as follows: we begin by examining the current state of immigrant health research, emphasizing the value of longitudinal methodologies in providing unique insights into the post-migration health of immigrants, and also highlighting the current barriers faced in conducting longitudinal immigrant health research in Canada. Next, we provide a brief overview of our literature search methodology, which identified 34 relevant peer-reviewed studies utilizing Canadian longitudinal datasets to examine health outcomes (both physical and mental health) and other lifestyle factors among immigrant and refugee populations. We then present key findings from the selected studies, discussing the major data sources used, their characteristics, the challenges faced, and the statistical methods employed. Finally, we provide a summary of our findings and present insights and methodological recommendations for future research.

2. Materials and Methods

2.1. Immigrant Health Dynamics and the Healthy Immigrant Effect

Several factors contribute to explanations of the Healthy Immigrant Effect (HIE). One theory suggests that healthier individuals, better equipped to endure the challenges of migration, are more likely to self-select for immigration, reinforcing this effect (Kobayashi et al. 2008; McDonald and Kennedy 2004). Another explanation is that the initial health advantage observed among immigrants partly results from the rigorous medical screening processes required for immigration, which prioritize individuals with fewer pre-existing health conditions (Kobayashi et al. 2008; McDonald and Kennedy 2004). However, over time, as immigrants assimilate into the host country, this health advantage tends to gradually diminish, with immigrants’ health aligning with or even exceeding the chronic disease risks of the native-born population, particularly in conditions like hypertension, diabetes, and cardiovascular diseases (Martínez 2013; Vang et al. 2017). This “immigrant paradox” has been consistently observed in both Canada and among other major immigrant-receiving countries (Kennedy et al. 2015; Sanou et al. 2014; Vang et al. 2017).

Studies have identified a variety of factors contributing to this decline, including cultural, socioeconomic, and psychosocial changes associated with the immigration process (Dogra et al. 2010; George et al. 2015; Murillo et al. 2015; Panchang et al. 2016; Sanou et al. 2014; Vang et al. 2017). Research has also examined various aspects that can significantly worsen immigrants’ health outcomes, such as acculturative stress, the underutilization of health services, and language barriers in accessing healthcare (Kalich et al. 2016; Murillo et al. 2015; Panchang et al. 2016; Salas-Wright et al. 2016). There is also a small but growing body of research examining the role of immigrants’ post-migration lifestyle changes, such as whether changes in their level of physical activity or alterations in their diet (e.g., immigrants’ post-migration dietary acculturation) contribute to the gradual health decline of immigrants over time in Canada (Mahmood et al. 2019; Ramos Salas et al. 2016; Kwon 2023; Aljaroudi et al. 2019).

With respect to the methodological approaches, despite the extensive body of research on immigrants’ health and well-being, existing studies are cross-sectional and rely on data collected at a single point in time. This approach also often relies on participants’ retrospective recall of past experiences or recent changes, which introduces bias and fails to capture the dynamic nature of health changes and transitions. Consequently, their findings often fall short of adequately addressing the long-term health implications of transitions over the course of immigrant integration in Canada.

2.2. The Significance of Longitudinal Methods in Understanding Post-Migration Health of the Immigrant Population

Longitudinal methods refer to specific research techniques used to collect and analyze data over multiple time points and are essential for the studying post-migration health of the immigrant and refugee populations (McDonald and Kennedy 2004; Newbold 2005a). Unlike cross-sectional studies, longitudinal research collects data from the same individuals at multiple time points, allowing researchers to observe how variables of interest change over time for the same subjects. This approach is particularly well-suited for studying health transitions and trajectories in the data, e.g., the onset of certain conditions, patterns of change in health status, or shifts in lifestyle behaviors, which often occur gradually and require data collection over time to be fully understood. Thus, longitudinal research can offer deeper insights into the long-term effects of immigration on health and other related mechanisms, making it crucial for capturing the nuances of health trajectories as immigrants adapt to new environments. Moreover, longitudinal research offers significant advantages in minimizing errors or biases in data collection compared to cross-sectional research.

Cross-sectional studies often require participants to recall past events or conditions in very short, highly controlled, or artificial environments. While this approach is effective for identifying associations at a given moment, it increases the risk of recall bias—errors or inaccuracies in remembering past events—and may limit the extent to which the data reflect participants’ natural real-world conditions and behaviors (Kip et al. 2001). In contrast, by collecting data at multiple intervals as they occur in natural settings, longitudinal research achieves higher ecological validity, meaning that the study’s methods, settings, and procedures more closely resemble participants’ natural environments and real-world conditions (Ghisletta and Spini 2004). By minimizing recall bias and enhancing ecological validity, longitudinal research provides data that is both temporally nuanced and applicable to real-life problems. This makes it particularly valuable for informing practical interventions and policies, such as those aimed at understanding immigrants’ post-migration health decline over time.

Despite its advantages, longitudinal research poses considerable operational challenges and costs, especially when applied to immigrant populations (Young et al. 2006; Twisk and Vente 2002). Tracking participants over extended periods demands substantial resources to address issues such as high mobility among immigrants, which can lead to elevated attrition rates or the loss of participants over time (Lipps et al. 2013). Factors contributing to this mobility include job opportunities, family reunification, and changes in residency status, making it difficult to maintain consistent contact with participants. Frequent relocations, updates to contact information, or even participants leaving the country can further exacerbate attrition, leading to missing observations and incomplete data (Vidal and Lersch 2021). This attrition poses risks to the validity of study findings, as the remaining participants may not represent the full diversity of the original sample. For instance, immigrants from low-income countries, who often face greater socioeconomic challenges, may be more likely to drop out of studies (Hoen et al. 2022), introducing potential biases into the results. Mitigating these challenges requires effective tracking methods, such as digital tools for maintaining contact, leveraging administrative data across agencies, fostering international collaborations, and applying advanced statistical techniques to handle missing data (Ibrahim and Molenberghs 2009; Nakai and Ke 2011).

In light of the operational challenges associated with longitudinal studies, some researchers have turned to the use of pooled cross-sectional data as a practical alternative (Ng 2019). This approach involves combining multiple independent cross-sectional datasets, where different individuals or groups are surveyed at different time points. Although technically not a longitudinal method, pooled cross-sectional data allows researchers to explore cohort effects (e.g., how health outcomes differ by immigration year) and to examine trends and transitions by comparing groups surveyed in different years. As a cost-effective and less logistically intensive alternative to longitudinal studies, this approach has been widely employed to address specific research questions in immigrant health, such as shifts in employment rates, health outcomes, or language proficiency over time. For example, recurring population health surveys (e.g., Canadian Community Health Survey or other Statistics Canada datasets) collected at regular intervals enable researchers to infer patterns of health adaptation or disparities between recent and established immigrants, providing valuable insights into macro-level trends.

However, pooled cross-sectional data cannot capture individual-level changes; instead, differences in group averages between time points serve as proxies for trends. This makes it difficult to observe within-person changes over time or identify patterns such as individual health trajectories, transitions (e.g., from healthy to unhealthy), and the effects of time-varying predictors. Furthermore, differences in sampling, data collection procedures, or unmeasured cohort effects between time points can introduce bias, complicating the interpretation of results. Two major sources of bias are cohort effects and compositional bias. Cohort effects (Rozenberg et al. 1990) arise when different groups of individuals—such as immigrants arriving in different decades—experience varying contexts, policies, or environments that influence their health outcomes. These differences may be mistakenly interpreted as time trends. Compositional bias (Backenroth et al. 2023), on the other hand, occurs when the composition of the sample shifts across survey years, distorting observed trends. For instance, if healthier immigrants are disproportionately represented in more recent surveys, this can create the illusion of improved health outcomes over time. These challenges highlight the complexities of pooling data from diverse sources, where heterogeneity and sampling differences may introduce bias (Backenroth et al. 2023).

2.3. Barriers to Longitudinal Immigrant Health Research in Canada: Data Availability

The limited amount of longitudinal research on immigrant health in Canada can largely be attributed to the lack of comprehensive datasets that track both “immigration” and “health” variables over time. Most of the existing studies on immigrant health have relied on cross-sectional survey data such as the Canadian Community Health Survey (CCHS) and the General Social Survey (GSS). These surveys provide valuable insights into various aspects of Canadian immigrants’ lifestyles and well-being, offering rich data on economic, social, and health outcomes. However, their cross-sectional designs are inherently limited in their capacity to capture the long-term health trajectories of immigrants and to further examine the mechanisms behind the gradual erosion of the HIE over time in Canada.

While several longitudinal data sources have been available in Canada, such as the Longitudinal Survey of Immigrants to Canada (LSIC) and the National Population Health Survey (NPHS), each has its own limitations that undermine their suitability for longitudinal immigrant health research. In addition to both datasets now being dated and no longer being updated, the LSIC primarily focuses on immigrants’ economic outcomes, offering limited variables on health outcomes and health lifestyle behaviors. Moreover, because the LSIC includes only immigrants, it lacks a Canadian-born comparison group. Conversely, the NPHS provides a Canadian-born reference group but focuses mainly on health outcomes, with limited data on immigrants’ social and economic integration. These limitations make it difficult to conduct comprehensive longitudinal analyses that explore the intersections of immigrant integration and health outcomes.

One approach to addressing the limited availability of longitudinal data for immigrant health research has been the development and use of administrative data. Administrative data, which includes records from government agencies and service providers, offers valuable information collected over time, making it well-suited for longitudinal research. Administrative data often covers large populations, allowing for robust statistical analyses and the examination of smaller subgroups, such as immigrants from specific regions or with particular socioeconomic characteristics. This capacity is particularly valuable in immigrant health research, where subgroup analyses can uncover nuanced patterns and disparities.

In Canada, Statistics Canada has taken proactive steps to enhance data availability by linking survey data with longitudinal administrative records. A notable example is the linkage between the Canadian Community Health Survey (CCHS) and the Longitudinal Immigration Database (IMDB). Such linkages help mitigate some of the challenges associated with the availability of longitudinal data in immigrant health research. Notably, this integration allows researchers to explore immigrants’ health trajectories while considering important contextual factors such as immigration status, admission category, and socioeconomic characteristics. By combining health survey data with immigration and tax records, the CCHS-IMDB linkage enables a more comprehensive an analysis of how immigrants’ health evolves over time and how it relates to factors like employment, income, and mobility. This approach helps to address several challenges associated with longitudinal research, such as the high cost of primary data collection, issues of participant attrition, and limited sample sizes in traditional longitudinal surveys. Furthermore, the CCHS is also linked to other administrative health databases such as the National Ambulatory Care Reporting System (NACRS), the Discharge Abstract Database (DAD), and the Ontario Mental Health Reporting System (OMHRS), which allows for enriched analyses that integrate self-reported health measures with data on health service use, hospitalizations, and mental health service encounters.

While data linkages offer significant potential, they also come with limitations. Data availability depends on the scope of linked datasets, and privacy concerns may restrict access to certain records. Moreover, administrative data may lack detailed health measures and self-reported information, making survey linkages essential for capturing a fuller picture of immigrants’ health experiences. Handling the linked data also requires researchers to be equipped with specialized skills to clean, integrate, and analyze complex datasets. Despite these challenges, the integration of administrative and survey data remains a promising solution for advancing longitudinal immigrant health research in Canada.

2.4. Literature Search Methodology

The aim of this literature search is to review existing peer-reviewed studies on the HIE, that utilized longitudinal methods to understand health conditions and health behaviors among immigrant and refugee populations in Canada. To ensure the robustness and validity of our search procedure, the search methodology was assessed and validated by the Social Sciences librarian at the University of Manitoba in June 2024. The librarian’s expertise in database selection, search term optimization, and methodological rigor enhanced the thoroughness and accuracy of our literature search. All search results were initially screened by reviewing the abstracts, and if the abstracts did not provide enough information, the methods sections were also reviewed to ensure they met our inclusion criteria.

The literature search was conducted across five major databases: PsycInfo, PsyArticles, PubMed, Scopus, and Google Scholar. Search terms were carefully selected to identify studies relevant to the immigrant and refugee populations, including terms such as “immigrant”, “foreign-born”, “newcomer”, and “refugee”. To capture a comprehensive view of health outcomes, the search included both physical and mental health-related keywords derived from previous research. Mortality was not included in our criteria. The search also encompassed health lifestyle behaviors, incorporating terms related to diet (e.g., “nutrition”, “diet”, “intake”, “consumption”, “eating”), physical activity (e.g., “physical activity”, “exercise”, “work out”), sleep, smoking, alcohol consumption, and healthcare access. Longitudinal studies were specifically identified in PsycInfo and PsycArticles using the “Longitudinal Study” filter and were found in Google Scholar through the keyword “longitudinal”. Since PubMed and Scopus did not yield additional relevant articles beyond those identified in other databases, they were not further searched for longitudinal studies in the expanded phase.

Articles were included in the final analysis if they met the following criteria: they were peer-reviewed, published in English, focused on immigrant and refugee populations in Canada, examined health status or lifestyle behaviors, and employed a longitudinal design, i.e., data were collected from the same individuals at multiple time points. We also excluded studies that used the term “longitudinal” in their titles or abstracts but did not actually employ longitudinal methods in their data analyses—such as those using data from only a single time point within a longitudinal database, or studies that used data collected at different time points but from different random samples rather than following the same cohort over time. This screening process produced in 34 peer-reviewed articles that met our specified criteria. Their key characteristics, including data sources, target populations, tracking periods, health outcomes, and analysis methods (longitudinal or not), are summarized in

Table 1. Data sources and methodological characteristics of selected studies (34 Articles).

Scheme	Data Source	Target Population and Sample Size	Tracking Period	Age at Baseline	Health Outcome	Analysis Method
Browne et al. (2017)	LSIC	Immigrants (n = 7055)	2000/2001–2004/2005	Mean age = 34.93	Emotional problems (self-report): measured as dichotomous (have emotional problems vs. no emotional problems) indicator of mental health.	Mixed-effects regression: generalized linear mixed models (for binary outcomes).
Calvasina et al. (2014)	LSIC	Non-refugee immigrants aged 18–60 (n = 2126)	2000/2001–2004/2005	20–29 = 28.9% 30–39 = 47.1% 40–49 = 18.2% >50 = 5.8%	Unmet dental care needs (self-report): measured as dichotomous (have unmet dental care needs vs. no unmet dental care needs since the last interview) indicator of dental care access.	Logistic regression: this study lacks statistical techniques tailored to handle the unique structure of longitudinal data, and it appears the authors were unaware of the need for such techniques.
Chen et al. (2010)	LSIC	Immigrants who were age 15 or older at the time of landing, and were employed four years post-arrival (n = 5215)	2000/2001–2004/2005	15–24 = 108 25–34 = 1336 35–44 = 964 45–54 = 229 >55 = 48	Perceived health status (self-report, 5-point Likert scale): categorized as dichotomous (fair/poor health vs. good/very good/excellent health) indicator of general health. Emotional problems (self-report): measured as dichotomous (have emotional problems vs. no emotional problems) indicator of mental health.	Logistic regression: The authors created binary outcomes to indicate a decline in health b/w Waves 1 and 3 and used these in logistic regression. No statistical method tailored to handle the unique structure of longitudinal data.
De Maio and Kemp (2010)	LSIC	Recent immigrants aged 15 and above (n = 7720)	2000/2001–2004/2005	15–24 = 1350 25–34 = 2880 35–44 = 2150 45–54 = 780 55–64 = 360 >65 = 200	Perceived health status (self-report, 5-point Likert scale): categorized as dichotomous (fair/poor health vs. good/very good/excellent health) indicator of general health. Emotional problems (self-report): measured as dichotomous (have emotional problems vs. no emotional problems) indicator of mental health.	Logistic regression: similar to Chen et al. (2010), refer to the Analysis Method column entry for Chen et al. (2010).
Fuller-Thomson et al. (2011)	LSIC	Recent immigrants aged 20–50 (n = 4684)	2000/2001–2004/2005	Mean age of full cohort (n = 7716) = 35.1 20–29 = 378 30–39 = 2971 40–49 = 1335	Decline or improvement in perceived health status (self-reported, 5-point Likert scale): categorized as dichotomous (two-step decline or improvement in health status vs. one-step or no change in health status) indicator of general health.	Logistic regression: Similar to Chen et al. (2010). Refer to the Analysis Method column entry for Chen et al. (2010).
Kim et al. (2013)	LSIC	Recent immigrants aged 20–59 (n =6660)	2000/2001–2004/2005	Mean age = 34.8 20–29 = 1921 30–39 = 2964 40–49 = 1335 50–59 = 440	Perceived health status (self-report, 5-point Likert scale): categorized as dichotomous (fair/poor health vs. good/very good/ excellent health) indicator of general health.	Generalized estimating squations (GEEs)
Kim and Noh (2015)	LSIC	Recent immigrants aged 15 and above from South Korea (n = 1181)	2000/2001–2004/2005	Mean age = 36 15–29 = 528 30–44 = 978 45+ = 310	Life satisfaction (self-report, 5-point Likert scale): categorized as dichotomous (completely satisfied/satisfied vs. dissatisfied/completely dissatisfied) indicator of well-being.	Logistic regression: Similar to Chen et al. (2010). The authors created a binary outcome indicating dissatisfaction with life b/w Waves 1 and 3.
Newbold (2009)	LSIC	Recent immigrants (n = not reported)	2000/2001–2004/2005	Not reported	Perceived health status (self-report, 5-point Likert scale): categorized as dichotomous (fair/poor health vs. good/very good/excellent health) indicator of general health.	Cox proportional hazards models to evaluate the risk of transitioning to poor health and hospitalization.
Robert and Gilkinson (2012)	LSIC	Recent immigrants (n = approximately 7700)	2000/2001–2004/2005	Not reported	Emotional problems (self-report): measured as dichotomous (have emotional problems vs. no emotional problems) indicator of emotional health. Stress (self-report, 5-point Likert scale): categorized as dichotomous (very/extremely stressful vs. not at all/not very/a bit stressful) indicator of stress level.	This study lacks methods tailored for longitudinal data, using logistic regression separately to each wave.
Setia et al. (2011a)	LSIC	Recent non-refugee immigrants aged 18 and above (n = 5082)	2000/2001–2004/2005	Mean age of: Males = 36 Females = 35.4	Perceived health status (self-report, 5-point Likert scale): categorized as dichotomous (fair/poor health vs. good/very good/excellent health) indicator of general health.	Multilevel random effect models
Shooshtari et al. (2014)	LSIC	Recent immigrants aged 15 and above from the Philippines (n = 529)	2000/2001–2004/2005	<25 = 13% 25–54 = 76% >55 = 11%	Perceived health status (self-report, 5-point Likert scale): categorized as dichotomous (fair/poor health vs. good/very good/excellent health) indicator of general health.	Logistic regression with predictors from Time 1 and Time 2 for outcomes at Time 3; the study lacks techniques tailored for longitudinal data.
Zhao et al. (2010)	LSIC	Recent immigrants aged 15 and above (n = 7700)	2000/2001–2004/2005	Mean age = 34	Perceived health status (self-report, 5-point Likert scale): categorized as dichotomous (fair/poor health vs. good/very good/excellent health) indicator of general health.	Generalized estimating equations (GEE)
Li et al. (2021)	NPHS	Adults aged 18 and above in Ontario (n = 29,838)	1996/1997–2006/2007	18–34 = 12,399 35–49 = 11,051 50–64 = 4579 >65 = 1809	Development of physical illness (diagnoses from the Institute for Clinical Evaluative Sciences database): categorized as dichotomous (have physical illness vs. no physical illness) indicator of physical health.	Cox proportional hazards models to assess the risk of developing physical illness and multimorbidity over a 10-year follow-up.
Newbold (2005a)	NPHS	Adults aged 20 and above (n = 1305 immigrants, native sample size not reported)	1994/1995–2000/2001	Mean age = 37.7	Perceived health status (self-report, 5-point Likert scale), healthcare use (self-report general practitioner contact, hospital use, alternative healthcare use, consultation with a healthcare professional): categorized as dichotomous (fair/poor health vs. good/very good/excellent health; use vs. non-use in the past year) indicator of physical health and healthcare use.	Cox proportional hazard model to assess the risk of transitioning to poor health and hospitalization over time.
Newbold (2005b)	NPHS	Adults aged 20 and above (n = 1305 immigrants, native sample size not reported)	1994/1995–2000/2001	Not reported	Perceived health status (self-report, 5-point Likert scale): categorized as dichotomous (fair/poor health vs. good/very good/excellent health) indicator of general health.	Cox proportional hazards models to assess the risk of transitioning from good to poor health over time
Newbold (2006)	NPHS	Adults aged 35 and above (n = 911)	1994/1995–2000/2001	Mean age = 51.65	Chronic conditions (self-report, 5 types—cardiovascular disease, asthma, arthritis, diabetes, any other conditions): categorized as dichotomous (presence of chronic condition vs. no chronic conditions) indicator of physical health.	Logistic regression to evaluate the presence of chronic conditions, Cox proportional hazard model to assess the risk of developing chronic conditions over time
Ng et al. (2005)	NPHS	Adults aged 18 and above (n = 14,117)	1994/1995–2002/2003	Not reported	Perceived health status (self-report, 5-point Likert scale), becoming frequent visitors to doctors (self-report), hospitalization (self-report): categorized as dichotomous (fair/poor health vs. good/very good/ excellent health; ≥6 doctor contacts a year vs. <6 doctor contacts a year; at least one hospitalization vs. no hospitalization) indicator of physical health and healthcare use. BMI (self-report height and weight): measured as continuous indicator of physical health	Discrete proportional hazards model (specified using a complementary log-log model, equivalent to Cox proportional hazards)
Pahwa et al. (2012)	NPHS	Participants 15 years and above (n = 14,713)	1994/1995–2004/2005	Not reported	Mental Distress Scale (self-rated, range 0–24): categorized as dichotomous (no/low distress [0–5] vs. moderate/high distress [6–24]) indicator of mental health.	Generalized estimating equations (GEEs)
Setia et al. (2009)	NPHS	Adults aged 18–54 (n = 5156)	1994/1995–2006/2007	Not reported	BMI (self-report height and weight): measured as continuous indicator of physical health	Random effects models
Setia et al. (2011b)	NPHS	Adults aged 18 and above (n = 7268)	1994/1995–2006/2007	Mean age of: Canadian born = 42.8 White immigrants = 50.9 Non-white immigrants = 39.8	Having a regular doctor and having unmet healthcare needs in the past 12 months (self-report): measured as dichotomous (have a regular doctor vs. no regular doctor, have unmet healthcare needs vs. no unmet healthcare needs) indicator of healthcare access	Random effects models
Setia et al. (2012)	NPHS	Adults aged 18 and above (n = 7268)	1994/1995–2006/2007	Not reported	Comprehensive International Diagnostic Interview (CIDI, self-report, range not reported) and perceived health status (self-report, 5-point Likert scale): categorized as dichotomous (CIDI ≥ 4 vs. CIDI < 4; fair/poor health vs. good/very good/excellent health) indicator of mental health and physical health	Random effects models
So and Quan (2012)	NPHS	Adults aged 18 and above (n = 9813)	1994/1995–2004/2005	18–39 = 2544 40–59 = 3018 >60 = 2472	Obese (calculate BMI from self-report height and weight), perceived health status (self-report), Healthy Utility Index Mark 3 (HUI3, self-report), and chronic condition (self-report, 8 conditions—heart disease, diabetes, kidney disease, HIV, high blood pressure, cancer, intestinal and stomach ulcers, dementia): categorized as dichotomous (BMI ≥ 25 vs. BMI < 25; fair/poor health vs. good/very good/excellent health; HUI3 ≥ median vs. HUI3 < median; have 1 of 8 chronic conditions vs. have none of 8 chronic conditions) indicator of physical health.	Multinomial logistic regression: The authors created a new categorical outcome by classifying changes in health status into three categories—improvement, decline, or no change over time. The study lacks statistical methods specifically tailored for longitudinal data.
Farid et al. (2020)	CLSA	Adults aged 45–85 (n = 23,002)	2012/2015–2015/2018	Mean age = 63 45–60 = 9866 61–70 = 6905 71–85 = 6231	Center for Epidemiological Studies Depression 10 (CES-D 10, self-report, range not reported): categorized as dichotomous (CES-D 10 ≥ 10 vs. CES-D 10 < 10) indicator of undiagnosed depression. Kessler Psychological Distress Scale 10 (K10, self-report, range not reported): categorized as dichotomous (K10 ≥ 19 vs. K10 < 19) indicator of depressive symptoms. Consulting a mental healthcare professional (MHCP, self-report): measured as dichotomous (consulted with a MHCP in 18 months or did not consult with a MHCP in 18 months) indicator of mental healthcare use.	Logistic regression to analyze variables at baseline and at the 18-month follow-up (no statistical analysis specifically designed for longitudinal data).
Farid et al. (2022)	CLSA	Adults aged 45–85 (Cohort 1 = 20,723 Cohort 2 = 22,054)	2012/2015–2015/2018	Cohort 1: Mean age = 62.7 45–60 = 9257 61–70 = 6479 71–85 = 4987 Cohort 2: Mean age = 62.1 45–60 = 10,492 61–70 = 6598 71–85 = 4964	Center for Epidemiological Studies Depression-10 (CES-D-10, self-report, range not reported): categorized as dichotomous (CES-D-10 ≥ 10 vs. CES-D-10 < 10) indicator of depression. Being treated for depression (self-report): measured as dichotomous (being treated for depression vs. not being treated for depression) indicator of depression. Glycated hemoglobin level (HbA1c, measured in physical examinations): categorized as dichotomous (HbA1c ≥ 7% vs. HbA1c < 7%) indicator of diabetes. Diabetes status (self-report): measured as dichotomous (have a doctor told them that they have diabetes, borderline diabetes, or high blood sugar vs. do not) indicator of diabetes.	Logistic regression to analyze variables at baseline and at the 18-month follow-up (No statistical analysis specifically designed for longitudinal data was conducted)
Ho et al. (2022)	CLSA	Successful aging adults aged 60 and above (n = 7651)	2012/2015–2015/2018	55–59 = 1164 60–64 = 2079 65–69 = 1708 70–74 = 1122 75–79 = 1018 >80 = 510	Physical wellness, psychological and emotional wellness, social wellness, and self-rated wellness (derived from multiple self-report scales and yes/no questions): categorized as dichotomous (have all four criteria vs. miss one criteria or more) indicator of successful aging.	Logistic regression (primarily focuses on cross-sectional comparisons; does not employ methods specifically designed for longitudinal data)
Khattar et al. (2023)	CLSA	Adults aged 45–85 (n = 23,972)	2012/2015–2020	Measured at COVID-19 baseline in 2020: 50–54 = 1097 55–64 = 7250 65–74 = 8759 75–84 = 5145 85–96 = 1721	Unmet healthcare needs (self-report): measured as dichotomous (“yes” to experienced challenges in accessing healthcare, did not go to hospital or to see a doctor when they needed to, and experienced barriers to accessing testing for COVID-19 vs. “no”/”don’t know” to experienced challenges in accessing healthcare, did not go to hospital or to see a doctor when they needed to, and experienced barriers to accessing testing for COVID-19) indicator of unmet healthcare needs during lockdown.	Logistic regression (primarily focuses on cross-sectional comparisons; does not employ methods specifically designed for longitudinal data)
MacNeil et al. (2024)	CLSA	Adults aged 45–85 with a stroke history (n = 577)	2012/2015–2020	Mean age = 74.56	Center for Epidemiological Studies Depression-10 (CES-D-10, self-report, range 0–30): categorized as dichotomous (CES-D-10 ≥ 10 vs. CES-D-10 < 10) indicator of depression. Diagnosis for depression (self-report): measured as dichotomous (have doctor told them that they had clinical depression vs. not have doctor told them that they had clinical depression) indicator of depression.	Logistic regression (it focuses on cross-sectional comparisons and does not employ longitudinal-specific techniques)
Dennis et al. (2018)	Original data	Chinese Canadian women postpartum (n = 571)	2011/2014–2012/2015	Mean age = 31.6	Edinburgh Postnatal Depression Scale (EPDS, self-report, range 0–30): categorized as dichotomous (EPDS > 9 for possible depressive symptomology or EPDS > 12 for high depressive symptomology vs. EPDS ≤ 9 or ≤12) indicator of depressive symptomology. State-Trait Anxiety Inventory (STAI, self-report, range 20–80): categorized as dichotomous (STAI > 40 vs. STAI ≤ 40) indicator for anxiety symptomology	Poisson regression with robust variance to estimate adjusted prevalence ratios for postpartum depressive and anxiety symptomatology. No specific longitudinal methods were employed.
Noh and Avison (1996)	KMHS	Korean immigrants aged 18 and above in Toronto (n = 609)	1990–1991	<36 = 25.5% 36–55 = 55.3% >55 = 19.2%	Center for Epidemiological Studies Depression (CES-D, self-report, range not reported): measured as continuous indicator of psychological distress.	Linear regression using depressive symptoms from the previous wave, along with other covariates, to predict symptoms in the next wave (a practice not recommended for longitudinal data analysis)
Quon et al. (2012)	NLSCY	Children and adolescents aged 6–17 (n = 26,442)	1994/1995–2006/2007	Mean age of: Children (6–11 years) = 8.67 Adolescents (12–17) = 14.07	BMI (self-report height and weight): categorized as percentiles (85th ≤ BMI < 95th—overweight; BMI ≥ 95th percentile—obese), indicator of physical health	Hierarchical linear modeling (Random effects models)
Salhi et al. (2021)	SYLS	Somali young adult refugees in North America (n = 383)	2013/2014–2014/2015	Mean age = 21.96	Hopkins Symptom Checklist (HSCL-25, self-report, range not reported): measured as continuous indicator of anxiety symptoms. Havard Trauma Questionnaire (HTQ, self-report, range not reported): measured as continuous indicator of PTSD symptoms.	Linear regression, separate models at Waves 1 and 2. No specific longitudinal methods were employed.
Saunders et al. (2018)	ICES	Youth aged 10–24 in Ontario (n = approximately 2.5–2.9 million)	1996/1998–2011/2012	10–14 = 45.6% 15–19 = 27.0% 20–24 = 27.4%	Outpatient physician visits, emergency department visits, hospitalizations for mental health related problems (billing data from other datasets): measured as continuous, as an indicator of mental healthcare use	Poisson regression using generalized estimating equations (GEEs)
Sou et al. (2017)	AESHA	Women sex workers aged 14 and above (n = 742)	2010–2014	Median age = 35 IQR = 28–42	Unmet health needs (self-report, 5-point Likert scale): categorized as dichotomous (always/usually receive healthcare services when needed vs. sometimes, occasionally, never, and N/A get healthcare services when needed) indicator of unmet health care needs.	Bivariate and multivariate logistic regression using generalized estimating equations (GEEs)
Walker and Ito (2017)	Original data	Mainland Chinese immigrants in Calgary (n = 115)	2 years (No exact start/end dates mentioned)	35–49 = 73%	Happiness and life satisfaction (self-report, 11-point Likert scale), Leisure Satisfaction Scale (LSS, self-report, range not reported): measured as continuous indicator of well-being.	Hierarchical linear modeling (Random effect models)

, Data sources and methodological characteristics of selected studies.

• The articles are ordered based on the data source that is most frequently used. Within each data source, the articles are presented in alphabetical order by the first author’s last name.
• In Li et al. (2021), participants were tracked for 10 years after their response to the NPHS (1996–1997) and CCHS (2000–2001, 2003).
• In Khattar et al. (2023) and MacNeil et al. (2024), the CLSA Comprehensive cohort at baseline (2011–2015), follow-up 1 (2015–2018), and two COVID-19 questionnaire waves (Spring 2020 and Autumn 2020) were used.
• Saunders et al. (2018) used linked health and administrative datasets in Ontario, Canada, spanning from 1996 to 2012.

3. Results and Discussion

3.1. Data Sources and Major Health Outcome Measures

The selected 34 articles employed a variety of datasets to explore the health trajectories of immigrants in Canada. These datasets are discussed in order of their frequency of use in the literature, beginning with the most commonly utilized: the Longitudinal Survey of Immigrants to Canada (LSIC), the National Population Health Survey (NPHS), and the Canadian Longitudinal Study on Aging (CLSA). We also briefly discuss other longitudinal datasets that appear less frequently in the literature.

3.1.1. LSIC, NPHS, and CLSA

The LSIC is the most frequently used dataset, featured in 12 studies (35% of the total). This survey, designed to examine how new immigrants adjust to life in Canada during their first four years of settlement, targeted immigrants who arrived between 2000 and 2001. Data were collected across three waves—at six months, two years, and four years post-arrival—and the survey was completed in 2005. This relatively short time frame allows for a closer examination of changes and adjustments made by immigrants shortly after their arrival in Canada. Covering key topics on adjustment and adaptation like language proficiency, employment, income, health, and social networks, this survey included around 12,000 participants, with data gathered through interviews in 15 languages, representing approximately 93% of the new immigrant population in Canada.

The studies included in our review that used this dataset were mostly published between 2009 and 2014 and focused on factors related to health deterioration among immigrants post-arrival. With the exception of one study (Calvasina et al. 2014), all studies used self-rated perceived overall health (ranging from excellent to poor) and self-rated emotional/mental health—where feelings of sadness, depression, loneliness, etc., were measured in a single question—as indicators of immigrant health. Key factors examined included immigration characteristics (e.g., ethnic background, immigrant class), socioeconomic status (e.g., income, employment, remittance behavior, job qualification alignment), as well as race/ethnicity-based discrimination and healthcare accessibility. The actual sample varied by target ethnic group, but generally centered around 7000 participants.

The NPHS is another widely employed dataset, used in 10 studies (29%). Initially, the NPHS included both cross-sectional and longitudinal components in its first three cycles (1994/1995, 1996/1997, and 1998/1999), but the survey became strictly longitudinal starting from Cycle 4 (2000/2001), following the same 17,276 individuals over time. Unlike LSIC, which specifically targeted immigrants, the NPHS tracks the health outcomes of the broader Canadian population every two years. The 2010/2011 cycle marks the final wave, resulting in a total of 9 waves of data. Among the cohort, less than 10% are white immigrants and less than 5% are non-white immigrants.

The studies using NPHS data were published between 2005 and 2021, but all utilized data were collected between 1994 and 2007, spanning up to seven waves. These studies examined immigrant health outcomes using a variety of measures: self-rated perceived overall health status, physical health (assessed by BMI or the presence/number of chronic conditions and physical illnesses), mental health (measured by levels of psychological distress), and healthcare access. Compared to the outcome measures used in the LSIC, the NPHS studies employed more multifaceted approaches. For example, mental distress was assessed using a well-developed six-item distress scale, and healthcare access was measured both through perceived measures (e.g., self-reported unmet healthcare needs) and more objective indicators (e.g., general practitioner contact, hospital use, alternative healthcare services, and consultations with healthcare professionals). With the exception of one study (Li et al. 2021), which combined NPHS from 1996 to 2006 with CCHS from 2000 to 2013 to provide a comprehensive view of physical illness development in adults relative to depression, most reviewed articles focused on basic demographic variables as influential factors, such as gender and ethnicity, or made broad comparisons between immigrants and non-immigrants. This reflects the limited immigration-related variables available in the NPHS. The actual sample sizes used in these studies ranged from a minimum of 911 to a maximum of 29,838, depending on the variables of interest.

The CLSA is another significant national longitudinal research, featured in five studies (15% of the total). The CLSA began in 2011 with baseline data collection from over 50,000 participants aged 45 to 85 (Wave 1). Follow-up 1 (Wave 2), conducted between 2015 and 2018, produced a 95% retention rate. Follow-up 2 (Wave 3), conducted between 2018 and 2021, is set to be accessible in early 2025, though its retention rate is currently unknown. Additionally, the CLSA carried out an ancillary study with a subset of around 28,000 participants, focusing on the impact of COVID-19. This study collected data online at two time points—April–May 2020 and September–December 2020—specifically targeting COVID-related health and behavior. The CLSA is designed to explore the diverse experiences and health transitions of aging Canadians, including those of immigrant origin, offering multidimensional insights into a wide range of health determinants—demographic, social, physical, clinical, psychological, and economic factors—along with biological specimens.

The five studies indicate that research using CLSA is gradually expanding, with all these articles published between 2020 and 2024. These studies address key concerns such as transitions in mental health, chronic conditions, and healthcare access, focusing on the intersection of these issues with immigrant status. The impact of COVID-19 pandemic-related healthcare challenges and mental health stressors among immigrants was highlighted in two studies (Khattar et al. 2023; MacNeil et al. 2024). Reflecting the aging cohort of the CLSA, one study focused on successful aging, revealing that immigrant older adults had lower odds of achieving successful aging compared to Canadian-born peers, especially when lifestyle and socioeconomic factors were considered (Ho et al. 2022). These areas, which were less comprehensively explored in older studies using datasets like the LSIC and NPHS, illustrate the expanding scope of research on immigrant health using the CLSA data. All five articles utilized data from the first two waves of the CLSA (Baseline and Follow-up 1, spanning 2011–2018).

3.1.2. Other Data Sources

Other population-based longitudinal datasets are used less frequently but still contribute to the overall understanding of immigrant health. For example, Quon et al. (2012) utilized data from the National Longitudinal Survey of Children and Youth (NLSCY), which tracked children and youth biennially from 1994 to 2007 across seven waves. Their study emphasized the significant impact of socioeconomic factors and family dynamics on health outcomes. Saunders et al. (2018) creatively linked multiple health and administrative databases, including the Immigration, Refugees, and Citizenship Canada (IRCC) Permanent Resident Database, which provided immigration status and demographic details, and health databases such as the National Ambulatory Care Reporting System (NACRS) and the Canadian Institute for Health Information Discharge Abstract Database (CIHI-DAD). This allowed scholars to track mental healthcare utilization in immigrant youth from 1996 to 2012. Their study found lower rates of mental health service use among recent immigrants compared to long-term residents, highlighting disparities in access to care.

To explore the unique health outcomes among specific ethnic groups and marginalized communities within the immigrant population, some articles relied on targeted datasets that provided a more focused examination of these populations. Their approaches involved targeted recruitment strategies tailored to the cultural and linguistic characteristics of their study populations. For example, Walker and Ito (2017) collected data over two years with four waves, recruiting Mainland Chinese Canadian immigrants from Calgary, using a list of the 100 most common Chinese surnames to ensure inclusivity. Eligible participants were those living in Canada for 10 years or fewer, with a near-equal balance of male and female participants, and data were collected via computer-assisted telephone interviews in English, Cantonese, and Mandarin to accommodate linguistic diversity. Dennis et al. (2018) tracked a community-based cohort in Ontario over three years (2011–2014) across three waves to investigate postpartum depression and anxiety among Chinese immigrant and non-immigrant women. Participants were recruited through public health home visitors, community organizations, and advertisements in Chinese-language media, ensuring accessibility for diverse populations. Telephone interviews in English, Cantonese, or Mandarin were conducted postpartum using validated instruments like the Edinburgh Postnatal Depression Scale (EPDS). Noh and Avison (1996) used data from the Korean Mental Health Study (KMHS), tracking data over a one-year period (1990–1991) with two waves, to examine the stress process and psychological distress among Korean immigrants in Canada. Additionally, Salhi et al. (2021) explored the impact of pre- and post-resettlement violence exposure on mental health among Somali refugees over a two-year period (2013–2015), collecting data in two waves from the Somali Youth Longitudinal Study (SYLS) dataset. Finally, Sou et al. (2017) utilized data from An Evaluation of Sex Workers Health Access (AESHA) to examine the correlates of unmet health needs and healthcare-related challenges among sex workers in Metro Vancouver, with a particular focus on immigrant women.

These studies are particularly valuable as they focus on specific groups not typically tracked in larger, population-based Canadian datasets. The datasets used in these studies have relatively short tracking periods, providing immediate insights into the health and well-being of immigrants shortly after resettlement or during significant life events. However, these short tracking periods also highlight the challenges of collecting longitudinal data within small research groups. Additionally, the focus on specific immigrant groups often results in smaller sample sizes (N = 100–700 in the reviewed articles), which can present difficulties in applying advanced statistical methods tailored to longitudinal data structures, potentially limiting the depth of analysis and generalizability of findings. This issue connects to the broader discussion in the following section, “Clarity in Reporting Dataset Usage and Studied Participants”, as achieving a high level of representativeness among targeted populations remains a challenge. For example, while Walker and Ito (2017) used a robust recruitment strategy to balance gender representation and capture variability across recent and more established immigrants, the first-wave response rate (26%) limited representativeness. Similarly, Dennis et al. (2018) ensured accessibility and cultural inclusivity by including multiple languages (English, Cantonese, Mandarin); however, exclusion criteria (e.g., mental health status) may have limited representation of individuals with higher vulnerability, a limitation the authors acknowledged.

3.2. Key Data Characteristics and Considerations for Longitudinal Research on Immigrant Health

This section explores the key features of major data sources that influence the scope, findings, and relevance of longitudinal research on immigrant health. It focuses on critical factors such as data accessibility, age distribution, temporal coverage, and alignment with the time period relevant to the research question. These factors are critical for contextualizing findings and avoiding overgeneralization, as they collectively influence the representativeness, generalizability, and depth of HIE research. By examining these aspects, this section highlights the strengths and limitations of existing datasets while underscoring the need for robust and inclusive data. Addressing these challenges is vital for advancing research and gaining a deeper, more comprehensive understanding of immigrant health trajectories.

3.2.1. Data Accessibility

Understanding data accessibility is crucial for researchers intending to use longitudinal datasets, as it significantly impacts the planning, execution, timeline, feasibility, and scope of their studies. The reviewed articles demonstrate three primary types of data access. First, Open Data with Attribution Requirements refers to datasets that are freely available to the public but require proper attribution or citation in any resulting publications. Examples include the Public Use Microdata Files (PUMFs) of the CCHS and the NLSCY.

Second, Controlled Access Data includes datasets that require approval from a data access committee or a similar governing body, typically involving the submission of a detailed research proposal and ethics approval before access is granted. Examples include the LSIC, NPHS, CLSA, AESHA, and the SYLS series. Some of these datasets are hosted and distributed through specialized data repositories such as the Inter-university Consortium for Political and Social Research (ICPSR), which facilitates secure data access and sharing across academic institutions. Additionally, Statistics Canada’s Research Data Centre (RDC) network provides access to more detailed versions of key datasets, such as CCHS, NLSCY, NPHS, LSIC, and many more. While RDC-based datasets offer enhanced analytical potential, they are subject to strict reporting and disclosure requirements to protect respondent confidentiality.

Lastly, some datasets are temporarily restricted to allow the original researchers or data creators an exclusive period to publish their findings before making the data accessible to others. Studies such as that of Walker and Ito (2017), Dennis et al. (2018), and Noh and Avison (1996) employed this type of original “primary” data, collected by the researchers themselves. These varying levels of data accessibility shape the depth, specificity, and scope of research outcomes, making it a critical consideration in study planning.

3.2.2. Age Distribution and Temporal Coverage

The age distribution of the studied population is a fundamental aspect of any health study, particularly in the context of immigrant health. It shapes the interpretation of findings, informs the development of interventions, and guides the formulation of policies to improve health outcomes across the life course (Bass et al. 2024; Bjerre et al. 2021). The average ages at baseline across the identified articles reveal significant variability, reflecting the populations with diverse ages studied in the current longitudinal HIE research. The LSIC has an average age of about 34 years, with participants ranging from their 20s to 40s, as seen in 11 articles included in our review. Similarly, two studies using the NPHS and one on Chinese immigrant women, with an average age of 31.6 years, fall within this age range. The SYLS, with a mean age of 23.96 years, and the AESHA dataset, with an average age of 35 years, also contribute to the understanding of younger, working-age immigrant populations. In total, 16 articles (47%) concentrated on this demographic. The CLSA, with participants ranging from their 50s to 70s, focuses primarily on older adults, highlighting its relevance to aging-related health outcomes. Studies utilizing the NLSCY (average age: 11 years) and the IRCC Permanent Resident Database (tracking individuals aged 10–24 years) addressed childhood and youth health. Notably, nine articles (26%) did not report specific age information (five using NPHS, one using LSIC, one using CLSA, and two using original data), highlighting gaps in reporting. Researchers should carefully consider the age group selected for their studies, as focusing on certain age groups may limit the generalizability of findings across the immigrant life course.

Temporal coverage of data collection is equally vital in shaping the findings of a longitudinal study. In the context of immigrant populations, the length of time spent in the host country is often closely related to the level of acculturation, which in turn affects health outcomes (Setia et al. 2012). Also, a longer data collection period enables researchers to observe how health outcomes change as immigrants adapt to their new environment, providing insights into the long-term impacts of acculturation, such as shifts in diet, social integration, and access to healthcare services (Newbold 2005a). In contrast, datasets like the LSIC, focusing on the initial four years of settlement, capture short-term adjustments, such as early employment challenges and initial healthcare access. While these shorter-term studies offer valuable immediate insights and reduce participant attrition risks associated with immigrant mobility, they may fail to capture broader, long-term trends. Such trends are essential for understanding the full spectrum of health challenges and opportunities within immigrant populations (Vang et al. 2017).

The interplay between age distribution and temporal coverage further defines the scope and quality of HIE research, as demonstrated by several reviewed studies. For instance, datasets focused on older populations, such as the CLSA, provide valuable insights into aging-related changes, including the progression of chronic conditions, shifts in healthcare utilization, and cognitive decline. However, these datasets often require longer follow-up periods to fully capture these trajectories. In contrast, studies utilizing datasets like the SYLS, with younger populations, prioritize the capture of developmental milestones, such as educational attainment and early employment integration, within shorter temporal windows. The alignment (or misalignment) of age distribution with temporal coverage determines whether datasets can effectively address the health challenges faced by specific subgroups. By thoughtfully considering and reporting on age distribution and temporal coverage, researchers can produce more comprehensive and impactful insights into the health trajectories of immigrant populations.

3.2.3. Relevance of Older Data and Limitations for Contemporary Immigrant Health Research

While the LSIC and NPHS datasets—both managed by Statistics Canada—have been invaluable in providing large, representative samples, the data used in the reviewed studies were collected nearly two decades ago (2000–2005 for LSIC and from the mid-1990s until 2007 for NPHS), capturing the health experiences of immigrants during those specific periods. At the time of their collection and initial use, these datasets represented the best available resources and contributed significantly to advancing the field of immigrant health research in Canada. However, given the significant changes in the racial/ethnic composition of immigrant populations over the past decades, along with shifts in health services, lifestyles, and social conditions in Canada, these datasets may not fully capture the challenges faced by today’s immigrants in the current context (Statistics Canada 2022, 2024). Further, almost 65% of the reviewed studies relied on these two datasets—potentially limiting the relevance of the findings to the current context of immigrant health research. In contrast, the CLSA, with data collection that began in the early 2010s, allows researchers to study the effects of more recent health policies, technological advancements in healthcare, and the specific challenges faced by the aging population, such as those brought about by the COVID-19 pandemic.

Although the CLSA was primarily designed for aging research, it is also suitable for immigrant health research, as it includes a substantial sample of individuals in the core “working age” range (45 years or older). The CLSA also contains many variables relevant to financial aspects—closely tied to the retirement of older Canadians—as well as a wide range of health-related variables. This makes it particularly useful for examining how aging intersects with immigrant status to influence health outcomes over time. However, it is important to note that the reviewed articles are currently based on the two available time points: baseline (2011) and Follow-up 1 (2015–2018). Once Follow-up 2 (2018–2021) becomes available, it could provide more valuable longitudinal insights into the health trajectories of immigrants.

3.3. Other Challenges Identified Regarding Data Sources

Several areas of improvement remain in order to enhance the comparability and transparency of future research.

3.3.1. Representation of Immigrants in NPHS and CLSA

It is important to note that while NPHS and CLSA are population-based, representative sample datasets that are readily accessible to researchers—unlike restricted original datasets—they were not specifically designed to focus on immigrant populations. Thus, unlike LSIC, NPHS and CLSA lack variables specifically tailored to the immigrant experience. This requires special attention when conducting immigrant health research using these data. According to 2021 Census, immigrants make up roughly 23 percent of the total Canadian population, yet both NPHS and CLSA have less than 15% of participants identified as immigrants, with the majority being white immigrants. Less than one-third of the immigrants in these datasets are non-white. This can result in insufficient statistical power to analyze immigrant-specific health outcomes and may not adequately capture the diversity of experiences within different immigrant racial/ethnic groups. Moreover, these datasets also do not distinguish between immigrant admission categories (e.g., economic, family, or refugee class). This limitation further constrains researchers’ ability to examine potential variations in health outcomes and service needs across different immigrant subgroups, despite the likelihood that these groups may experience distinct settlement trajectories and health challenges.

In this aspect, studies focusing more on the heterogeneity of the immigrant subgroups, despite not always being publicly available, can provide deeper insights into the unique challenges and health outcomes faced by different immigrant subgroups. While the reviewed studies using such data may be limited in their generalizability due to smaller sample sizes and limited time waves, they contribute significantly to our understanding of subgroup differences within the immigrant population. Encouraging and supporting such research is essential for building a more nuanced and comprehensive understanding of immigrant health.

Furthermore, as the CLSA continues its data collection, there is an opportunity to enhance its utility for immigrant health research by incorporating additional variables that capture key aspects of the immigrant status and experience, such as length of stay in the host country, age at the time of first arrival, immigration generation (e.g., first or second generation), and language proficiency in the official languages of the host country at the time of arrival. Since these variables are time-invariant—meaning they do not change over time—they can be effectively integrated into the current longitudinal cohort data. Including these elements would significantly enhance the ability of the CLSA to support more immigrant-specific health research.

To our knowledge, the CLSA is the only longitudinal data that is currently being conducted as a large-scale, national, long-term study/platform. This means having a good representation of the current Canadian population is important, as there is a wide range of health and aging-related research across many disciplines that rely on the CLSA to understand these issues. Incorporating a few additional time-invariant variables, as suggested, is relatively straightforward. However, achieving a cohort that meets a good representation of the population can be challenging, especially for longitudinal data that has been collected for over 10 years. Therefore, one of the most significant challenges faced by researchers in immigrant health is the difficulty of obtaining robust, representative longitudinal data.

3.3.2. Underutilization of Administrative Health Data

Only one study (Saunders et al. 2018) from our literature search utilized administrative health data (e.g., hospital records, health insurance claims) despite its potential to provide objective, comprehensive data on healthcare utilization, disease incidence, and treatment outcomes. Administrative data is often underutilized in exploring the long-term health trajectories of immigrants. The greater use of administrative health data could address gaps in understanding how healthcare access and utilization impact long-term health outcomes (Lucyk et al. 2015; McKay et al. 2022). It could also provide more accurate and detailed information on chronic disease progression and the effectiveness of healthcare interventions among immigrant populations (Chiu et al. 2016).

Our initial literature search revealed that there are several studies that have used such data for health research and/or immigration research. But when it comes to utilizing Canadian administrative data in conducting longitudinal research specifically focused on “immigrant health”, the Canadian Community Health Survey (CCHS)–The Longitudinal Immigration Database (IMDB) linkage is currently one of the most viable administrative data sources available. Nonetheless, many studies using the CCHS-IMDB linkage were excluded from our review because they did not meet our strict search criteria, especially with respect to the methodological approach used. Several studies used this “longitudinal data” without employing true longitudinal methods for their analyses. Instead, they applied statistical techniques typically used for cross-sectional analyses and presented data from multiple time points as trends over time, rather than using longitudinal statistical techniques. For example, a recent study by Antonipillai et al. (2021) used the CCHS-IMDB-linked data but applied multivariable logistic regression to estimate the association between explanatory variables and binary health service utilization outcomes. Therefore, even when studies used linked data such as the CCHS-IMDB, it was essential for us to carefully evaluate each article to assess the appropriateness of the statistical methods applied in the analyses. Additionally, the CCHS-IMDB linkage has a notable limitation, as it includes data from only two focus content cycles: the 2002 and 2012 CCHS Mental Health and well-being surveys. These surveys gathered information on mental disorders, the use of health services related to mental health, and disabilities related to mental health among household residents aged 15 and older. As a result, research on immigrant health using this linkage remains largely confined to mental health domains, with limited exploration of other health-related factors such as dietary quality, physical activity participation, and overall lifestyle behaviors.

3.3.3. Clarity in Reporting Dataset Usage and Studied Participants

Another significant area for improvement identified in the review of selected articles is the clear reporting of the original data source and the specific participants analyzed. Conclusions should be firmly based on the actual sample included in the study, adhering strictly to the inclusion or exclusion criteria established for the research focus or statistical procedures. However, some articles tend to overextend their discussions and conclusions to a broader population than the sample analyzed justifies. This often occurs when authors rely on the general characteristics of the original data source rather than the specific attributes of the analyzed sample (e.g., Farid et al. 2022; Ho et al. 2022; Khattar et al. 2023; Kim et al. 2013). For example, Farid et al. (2022) analyzed subsets of CLSA participants, excluding individuals with baseline diabetes or depression. Despite this restricted sample, the study’s conclusions broadly discuss bidirectional associations and implications for public health surveillance in both immigrant and non-immigrant populations. These generalizations are based on assumptions about the broader characteristics of CLSA participants rather than being limited to the analyzed subsets. Such practices risk interpretations that are not fully aligned with the data used in the statistical analyses, underscoring the need for more precise reporting and interpretation. Furthermore, a review of the major population-based data sources reveals varying sampling strategies across datasets. Unfortunately, these sampling strategies are not always fully addressed or acknowledged in the reviewed articles. These differences should be carefully considered when interpreting and generalizing findings. By providing clearer details on participant criteria, sampling strategies, and the scope of their conclusions, researchers can enhance the transparency, accuracy, and impact of their studies. Greater attention to these best practices will contribute to more robust and reliable findings in HIE research.

3.4. Statistical Procedures and Their Implications

3.4.1. Considerations in Outcome Simplification and Measurement Invariance

In the reviewed studies, three primary considerations regarding outcome variables were observed. First, a common practice was the dichotomization of continuous outcomes, often without detailed justification of the chosen threshold values. This is detailed in the column titled “Health Outcome Measures” in Table 1 for each article. While dichotomization can simplify interpretation and enhance the clarity of results, it comes with trade-offs, including the loss of information, the introduction of biases in estimating effects, and reduced statistical power, e.g., see Yoo (2010). By converting a continuous variable into two categories, the analysis focuses solely on the difference between these categories, rather than utilizing the full distribution of the original variable. This is particularly problematic when one category is dominant over the other, as it may not adequately represent the original distribution. Such simplification can result in the less efficient estimation of effects, potentially leading to non-significant results, even when a meaningful association exists. The choice of cutoff points is therefore critical, as it directly impacts both analysis and interpretation. The decision must be supported by sound scientific reasoning, such as a thorough literature review or statistical justification (e.g., conducting sensitivity analysis).

The second consideration involves the creation of new outcome measures by authors to reflect temporal change, rather than directly using the original data. Five articles (15% of the total) employed this approach, as noted in the “Analysis Methods” column of Table 1. For instance, Chen et al. (2010) used self-reported overall health status (on a 5-point Likert scale) and self-reported emotional problems (binary: presence or absence), collected at two time points, as their primary outcome measures. The authors then created new binary outcomes to indicate a decline in health status between these time points (decline or no decline), which were used in logistic regression. This practice, beyond concerns about dichotomization, introduces additional risks of misinterpreting changes over time. Specifically, it results in a loss of valuable information about the timing and magnitude of changes, leading to the less accurate representation of longitudinal data. For example, treating a decline from “excellent” to “good” the same as a decline from “fair” to “poor” may not accurately reflect the severity of health changes over time. Collapsing time points in this manner can obscure the conceptual clarity of the measures, making it unclear whether they reflect meaningful clinical changes or are simply statistical artifacts. Moreover, such simplification can mask individual differences in health trajectories that could be more effectively analyzed if the data were preserved in its original longitudinal form.

The third consideration relates to the predominant use of self-rated health (SRH) measures as primary outcomes. While these measures are valuable for capturing subjective health perceptions—and are often the only available indicator of general health status in large-scale datasets—they also present methodological challenges that warrant attention in longitudinal immigrant health research. A large body of evidence supports the validity of SRH, demonstrating strong associations with morbidity, mortality, and chronic conditions. However, SRH is also prone to differential item functioning (DIF), where participants from different backgrounds may interpret survey items differently. This variability can lead to inconsistencies in how outcomes are understood across diverse groups. In longitudinal studies, this challenge is further complicated by the response shift (RS), where participants’ perceptions or interpretations of the same items change over time due to shifts in internal standards, values, or conceptualization of the measured constructs. Both DIF and RS are closely related to the issue of measurement noninvariance, where a construct may not retain the same structure or meaning across different groups or measurement occasions. This noninvariance poses challenges for meaningful comparisons, as it suggests that the construct cannot be consistently interpreted across contexts or over time. Researchers must acknowledge these issues and adopt strategies to address them. The body of research on Patient-Reported Outcome Measures (PROMs) provides useful guidance for tackling the complexities of DIF and RS in self-rated health measures, offering valuable insights for researchers aiming to ensure more accurate and meaningful longitudinal comparisons.

3.4.2. Participant Selection and Handling of Missing Data

In most of the articles reviewed, the decisions about which participants to include (or exclude) were primarily driven by the technical needs of data management and analysis, rather than a strict focus on selecting participants who fully represent the target population. For instance, many studies selected participants based on whether they had complete observations at all time points or data available for key study variables, thereby prioritizing data completeness over population representativeness. While this simplifies data analysis by reducing the need for extensive missing data management, it can introduce biases and limit the generalizability of findings to the target immigrant population.

Technically, this procedure implicitly assumes that missing data are completely random and noninformative—that is, missing completely at random (MCAR)—and can therefore be ignored or deleted, which is unreasonable in most situations (Liu 2016). Ignoring complex or non-random missing data patterns when applying statistical methods can result in biased parameter estimates and erroneous predictions (Little and Rubin 2002). Notably, only a few articles (e.g., Quon et al. 2012; Setia et al. 2009, 2011a, 2012) clearly addressed the impact of participant exclusion due to missing observations on their statistical findings and overall conclusions, as well as how they handled missing data. The lack of such discussions creates a significant gap in understanding how the exclusion of participants affects the interpretation of results and the applicability of findings to the broader population.

Accurately addressing missing data in a statistical model depends on correctly identifying the underlying missing data mechanism, typically classified into three categories: missing completely at random (MCAR), missing at random (MAR), or missing not at random (MNAR). However, there is no statistical procedure that can definitively select the correct missing data mechanism, as the full-data models fitted to incomplete data are inherently non-identifiable (Liu 2016). In other words, when data is incomplete, the reason for the missingness—whether MCAR, MAR, or MNAR—cannot be conclusively determined using statistical methods alone. This challenge underscores the importance of incorporating relevant theoretical insights about variable relationships to better understand the distribution of missing data. By applying domain expertise and drawing on existing literature, researchers can make more informed assumptions about the missing data mechanism, leading to better-informed decisions in data analysis.

3.4.3. Sample Attrition in Longitudinal Research

Closely related to the issue of missing data is sample attrition, which occurs when participants drop out or are lost to follow-up over time. Attrition is a common and critical challenge in longitudinal studies, particularly in immigrant health research, where participants often experience high levels of mobility due to employment changes, family reunification, or shifts in residency status (Lipps et al. 2013; Vidal and Lersch 2021). These dynamics make consistent follow-up difficult and increase the likelihood of missing observations over time (Hoen et al. 2022).

Despite its importance, sample attrition was rarely addressed in the reviewed studies as a distinct methodological concern. Most articles did not report on attrition rates or examine whether dropout was associated with participant characteristics, nor did they evaluate the potential impact of attrition on study findings. When participants lost to follow-up differ systematically from those who remain—for example, if those experiencing more severe health or economic challenges are more likely to drop out—the resulting estimates may be biased, compromising both internal and external validity.

Several methodological approaches are available to address attrition-related bias. The Heckman selection model offers a two-step procedure that first models the probability of being included in the final analytic sample, and then estimates outcome models while accounting for potential selection bias (Heckman 1979). While this approach has not been widely adopted in the studies reviewed, it provides a useful framework for addressing non-random attrition (Liu 2016). Other strategies include sensitivity analyses, where comparisons are made between models using complete-case data and those using imputed or weighted data (Molenberghs and Kenward 2007; Carpenter et al. 2023), and inverse probability weighting, which reweights observed participants based on the probability of remaining in the study (Seaman and White 2013). These techniques rely on observed covariates to adjust for differential attrition, though they cannot fully correct for unmeasured sources of bias.

Given that attrition is often unavoidable in longitudinal research, especially with immigrant populations, researchers should routinely report attrition rates, explore predictors of dropout, and interpret findings in light of potential selection effects. Acknowledging and addressing attrition strengthens the transparency, validity, and policy relevance of immigrant health research using longitudinal data.

3.4.4. Common Statistical Methods

This section briefly reviews how each article utilized longitudinal data in their analyses (see Table 1 for details) and provides a concise overview of key statistical methods for educational purposes, focusing on their application in longitudinal research.

A prevalent approach involves using regression techniques, such as linear or logistic regression, which are traditionally designed for cross-sectional data. Approximately 50% of the reviewed articles employed these models in one of three ways, as described earlier: (1) in independent models operating at separate time points (i.e., treating the data as cross-sectional), (2) by using predictors or covariates from one time point to predict outcomes at a subsequent time point, or (3) by employing newly created outcome measures. These analytical approaches leverage the advantages of regression, including the ease of interpretation, the ability to manage various covariates, and the capacity to examine interactions among covariates.

However, linear and logistic regression models do not fully account for the longitudinal nature of the data, as they are limited in their ability to account for how outcomes change over time. Longitudinal data typically captures the evolution of an individual’s health or other characteristics over multiple time points, and these regression methods do not model this temporal aspect. This limitation also affects their inability to capture intra-individual variability. Individuals often follow different trajectories over time, and longitudinal data is uniquely suited to answering questions about these variations. As a result, statistical techniques such as mixed-effects models or growth curve models are better suited for capturing individual differences in these trajectories.

Additionally, when regression models are applied to newly created outcome measures—such as change scores or difference scores between time points—care is needed in interpreting the resulting coefficients. For example, a positive coefficient for a covariate may suggest that individuals with higher values on that covariate tend to show greater improvement over time. However, this interpretation applies only to the constructed change variable and does not necessarily reflect how the covariate relates to the original outcome at each time point. A coefficient near zero may indicate little or no association with the change score, but this does not imply that the covariate has no meaningful relationship with the underlying health measures themselves. Given that nearly 50% of the current literature relies on these regression techniques, interpretations of these findings should be approached with caution.

Survival analysis, particularly through the Cox proportional hazards model, is a widely utilized technique designed to analyze time-to-event data. This method focuses on estimating the time until a specific event occurs, such as the onset of a disease, recovery, or death. The Cox model operates by modeling the hazard function, which represents the risk of the event occurring at a specific time point, while simultaneously accounting for covariates that may influence this risk. Typically, covariates (e.g., age, BMI, smoking status) are collected at baseline, and the occurrence of the event is tracked over time. A significant strength of survival analysis is its capacity to handle censored data, which refers to situations where the exact timing of the event is unknown for some participants, such as due to dropouts or the event not occurring within the study period. While survival analysis is well-suited for studying time-to-event outcomes, its applicability hinges on the assumption of proportional hazards. This assumption stipulates that the effect of each covariate on the hazard remains constant over time. When this assumption is violated—such as in cases where covariate effects evolve dynamically over the course of a study—the results of the analysis may be biased, leading to incorrect inferences. This limitation is particularly relevant in longitudinal studies, where the influence of predictors often varies over time, reflecting the evolving nature of individual trajectories. Such violations necessitate careful diagnostic checks and, in cases where the assumption does not hold, the adoption of alternative modeling approaches is necessary.

Another widely used method in the reviewed studies is mixed-effects modeling, employed in 21% of the articles (7 articles), particularly in studies with short repeated-measures designs, typically involving two or three time points. Mixed-effects models—also known as random effects models, multilevel models, or hierarchical models—refer to the same underlying statistical approach.. Unlike the regression approaches mentioned earlier, mixed-effects models enable the direct modeling of repeated outcomes without requiring variable creation or transformation. These models maintain a similar structure to traditional regression, where relationships between covariates and outcomes are modeled, but they also account for the correlation between repeated measures within individuals by specifying “random effects”. That is, random effects generally account for variations due to the data’s grouping or nesting structure; thus, in longitudinal settings, this reflects within-individual changes over time. Random effects capture within-individual changes over time, while fixed effects capture population-level trends, making mixed-effects models particularly useful for understanding both within- and between-individual variability. Furthermore, mixed-effects models have other advantages, including the ability to accommodate unbalanced data and handle missing data points more effectively than simpler approaches, making them particularly useful in complex longitudinal designs.

Despite their strengths, mixed-effects models come with challenges, particularly in terms of model specification and interpretation. The complexity increases when multiple variables are considered to have random effects. Mixed-effects models incorporate both fixed effects (effects that are constant across individuals, such as overall population trends) and random effects (individual-specific variations), and determining which variables should have random effects requires careful consideration. This complexity can lead to the need for the use of various model comparison procedures to identify the best-fitting model. Additionally, as the complexity of random effects structures increases, models are more likely to face convergence issues (i.e., failure to produce valid statistical inferences). Estimating random effects in complex models often requires large sample sizes to achieve reliable results. Another challenge lies in specifying the correct correlation structure. As the complexity of the data increases, so does the difficulty of modeling the correlations between repeated measures appropriately. Consequently, the number of covariates or interactions considered in a mixed-effects model is often limited.

Finally, generalized estimating equations (GEEs) offer an alternative when the correlation structure of repeated measures is not of primary interest, but retain some characteristics of mixed-effects models. GEEs were applied in 15% of the reviewed articles (5 articles). They are particularly useful for estimating overall population trends rather than individual-level changes, as they focus on the average effects of covariates across the population. GEEs use a working correlation structure that may differ from the true correlation among the repeated measures, but still produce statistically robust results. This makes GEEs particularly useful when the primary focus is on estimating the regression relationship between covariates and the repeated measures due to their computational simplicity and robustness to the misspecification of the correlation structure.

4. Conclusions

This review provides a comprehensive examination of the current state of longitudinal research on immigrant health in Canada, with a particular focus on studies investigating the health trajectories of immigrant and refugee populations. Despite the acknowledged value of longitudinal studies in understanding health dynamics over time, we identified only 34 peer-reviewed articles that utilized Canadian longitudinal datasets to explore physical and mental health outcomes in these populations.

We conducted a detailed analysis of the identified articles, providing a comprehensive review of the key data sources, including the LSIC, NPHS, and CLSA. We also highlighted significant challenges, such as the limited availability of immigrant-specific variables and the underrepresentation of non-white immigrants. Additionally, many of these studies relied on data collected during earlier periods (e.g., LSIC: 2001–2005; NPHS: 1994–2009), which may not fully capture the health experiences of contemporary Canadian immigrant populations, given the significant changes in socio-demographic characteristics and immigration patterns in Canada over the past decades.

In addition to these data-related limitations, our review highlights several methodological concerns across the selected studies, such as the prevalent use of self-reported health measures prone to DIF and RS, especially in longitudinal contexts. We also noted the frequent oversimplification of outcome variables, which reduces the richness of insights that longitudinal data can offer. Although some studies have applied statistical methods that align with the longitudinal design of the data, many continue to rely on cross-sectional techniques, which fail to take full advantage of the temporal dimension that longitudinal studies provide. This underutilization of the longitudinal nature of the data and the simplification of analytical approaches hampers the potential of longitudinal studies to provide a more comprehensive view of immigrant health changes over time.

To our knowledge, this is one of the few to critically assess the use of longitudinal methods in Canadian immigrant health research. By offering this comprehensive review, we aimed to clarify the current state of the literature, provide insights into best practices for using longitudinal data in HIE research, and offer methodological suggestions for future studies. Our motivation for conducting this review stems from the critical need to enhance our understanding of how immigrant health evolves over time, which is essential for informing policies that support the long-term well-being of these populations. Longitudinal methodologies are particularly valuable for integrating theoretical frameworks, such as the life course approach, which emphasizes critical transitions and “turning points” that shape post-migration health outcomes (Castañeda et al. 2015; Terragni et al. 2014). By capturing these pivotal moments—whether related to migration, shifts in financial circumstances, or the changes in living conditions—longitudinal studies provide a more nuanced understanding of how immigrant health trajectories are influenced by evolving social, economic, and cultural contexts (Elder et al. 2003). Extending this further to incorporate an intersectional life course perspective could further provide deeper insights into how dimensions such as gender, class, and race/ethnicity intersect with immigration status to shape health outcomes (Bowen et al. 2023; Mycek et al. 2020).

Our analysis of the methodological practices across the reviewed studies underscored the importance of designing more rigorous and transparent longitudinal research. The findings from this review highlighted the need for updated, more representative datasets and the application of statistical techniques tailored to longitudinal designs to fully capture the complexities of immigrant health trajectories. While sophisticated longitudinal methods (e.g., latent growth curve modeling, dynamic structural equation models, longitudinal mediation models, and multilevel autoregressive modeling, etc.) are well-established and commonly applied in other fields like psychology, epidemiology, and related disciplines, their application in HIE research remains relatively limited.

Moving forward, fostering interdisciplinary collaboration—particularly with experts in complex data structures and immigrant health—could significantly enhance the methodological rigor and overall quality of Healthy Immigrant Effect (HIE) research. This review aimed to make these discussions more accessible to researchers and practitioners, including those with limited statistical expertise. This offered a valuable resource for a wide range of stakeholders involved in immigrant health research in Canada, including academics, policymakers, and community organizations. Through its accessible approach, the review sought to provide stakeholders with methodological insights to allow more meaningfully engagement with HIE research, fostering a collaborative environment where interdisciplinary insights could drive innovative solutions to immigrant health challenges.