Cross Sectional and Case-Control Study to Assess Time Trend, Gender Differences and Factors Associated with Physical Activity among Adults with Diabetes: Analysis of the European Health Interview Surveys for Spain (2014 & 2020)

(1) Background: We aim to assess the time trend from 2014 to 2020 in the prevalence of physical activity (PA), identify gender differences and sociodemographic and health-related factors associated with PA among people with diabetes, and compare PA between people with and without diabetes. (2) Methods: We conducted a cross-sectional and a case–control study using as data source the European Health Interview Surveys for Spain (EHISS) conducted in years 2014 and 2020. The presence of diabetes and PA were self-reported. Covariates included socio-demographic characteristics, health-related variables, and lifestyles. To compare people with and without diabetes, we matched individuals by age and sex. (3) Results: The number of participants aged ≥18 years with self-reported diabetes were 1852 and 1889 in the EHISS2014 and EHISS2020, respectively. The proportion of people with diabetes that had a medium or high frequency of PA improved from 48.3% in 2014 to 52.6% in 2020 (p = 0.009), with 68.5% in 2014 and 77.7% in 2020 being engaged in two or more days of PA (p < 0.001). Males with diabetes reported more PA than females with diabetes in both surveys. After matching by age and gender, participants with diabetes showed significantly lower engagement in PA than those without diabetes. Among adults with diabetes, multivariable logistic regression showed confirmation that PA improved significantly from 2014 to 2020 and that male sex, higher educational level, and better self-rated health were variables associated to more PA. However, self-reported comorbidities, smoking, or BMI > 30 were associated to less PA. (4) Conclusions: The time trend of PA among Spanish adults with diabetes is favorable but insufficient. The prevalence of PA in this diabetes population is low and does not reach the levels of the general population. Gender differences were found with significantly more PA among males with diabetes. Our result could help to improve the design and implementation of public health strategies to improve PA among people with diabetes.


Introduction
Diabetes is a prevalent chronic disease with a significant impact on quality of life, morbidity, mortality, and healthcare costs [1,2]. The 10th Edition of the Diabetes Atlas of the International Diabetes Federation estimates that 537 million adults are living with diabetes, with a continuous increase in prevalence, making this disease one of the main in our investigation, we have matched people with and without diabetes by age, gender, and region of residence to improve study efficiency by increasing precision and therefore providing novel and more reliable results [32].
The results or our investigation will provide policymakers with relevant data on time trends in PA and valuable information to target promotion and educational interventions to improve PA for those population groups with diabetes that would benefit most. Beside the reductions in the morbidity and mortality that increasing PA would yield in people with diabetes, a recent Spanish investigation estimated that EUR 2151 per individual may be saved if a minimum level of PA is implemented, due to a decrease in absenteeism and a lower use of healthcare services) [2,6,9,16].
In our opinion, all the mentioned issues require more investigation.
Using two EHISS conducted in years 2014 and 2020, the objectives of our investigation were to (i) assess the temporal trend in self-reported PA among people with diabetes from 2014 to 2020; (ii) identify gender differences in the frequency of PA among people with diabetes; (iii) compare the frequency of PA between people with diabetes and gender-agematched non-diabetic subjects; and (iv) determine which sociodemographic and healthrelated variables were associated with reporting PA in people with diabetes.

Study Design and Data Source
To reach the proposed objectives, we have conducted a cross-sectional and a casecontrol study. The data source was two EHISS corresponding to years 2014 (EHISS2014) and 2020 (EHISS2020).
Details on the EHISS2014 and EHISS2020 are available online [12,13]. Both surveys have identical methodology and questions [12,13]. Briefly, the EHISS is a home-based personal interview conducted with a three-stage sampling method to obtain a national representative sample of people aged ≥15 years residing in households.
The EHISS2014 was conducted from January to December 2014 and the EHISS2020 from July 2019 to July 2020. Due to the COVID-19 pandemic, during the last months (March to July) of the EHISS2020, the interviews could not be fulfilled at the person's home, so they were conducted by telephone [13].

Study Population and Matching Method
The study populations included all adults (≥18 years) interviewed in the EHISS2014 and the EHISS2020. A participant was considered to have diabetes if answered affirmatively to the question: "Has your doctor told you that you are suffering from diabetes?". Those who answered "no" were classified as non-diabetic subjects.
For each person with diabetes "case", we randomly matched a person without diabetes "control" interviewed in the same year and with identical age, gender, and region of residence.

Study Variables
Two dependent variables have been created to assess how frequently participants engaged in PA. To do so, we used two questions. The first question used was "Which of these possibilities best describes how often you do some PA in your free time?", with four possible answers: 1. "I don't exercise. I occupy my free time almost completely sedentary"; 2. "I do some occasional physical or sports activity"; 3. "I do PA several times a month"; and 4. "I do sports or physical training several times a week". With this question, we created the variable "Frequency of PA" and those participants who answered options 1 and 2 were classified as "sedentary or with a low frequency", and those who answered options 3 and 4 were classified to have a "medium or high frequency" of PA.
The second question used was "How many days, in a typical week, you do sports, gymnastics, bicycling, or walking fast for at least 10 min continuously?". The possible answers (0 to 7 days) were categorized in "none or one day" and "two days or more". This study variable was named "Number of days per week of PA". For study purpose, we excluded those patients who did not complete these questions or answered "I don't know". Sociodemographic covariates included gender, age, educational level, and if the person interviewed lived with a partner or not. Details of the questions used and the covariates created are shown in Table S1.
The health-related covariates analyzed were self-rated health over the last year and selfreported presence of physicians diagnosed chronic conditions. Conditions included chronic obstructive pulmonary disease (COPD), heart diseases, stroke, cancer, mental disease, and high blood pressure. Information regarding lifestyles such as alcohol consumption, active smoking, and body mass index (<25, 25-29.9, and ≥30) were also collected, as detailed in Table S1.

Statistical Analysis
The frequency of PA and the number of days per week of PA was estimated according to study covariates for people with (cases) and without diabetes (controls).
Absolute numbers with percentages are shown for qualitative variables and means with standard deviations for quantitative variables. To compare unmatched qualitative variables, we used the chi-square test. We checked the normality of continuous variables using the Kolmogorov-Smirnov test and found that our study variables followed a normal distribution, so the Student's t-test was used for comparisons. For matched comparison, the corresponding tests applied were McNemar's test and paired Student's t-tests as required.
Multivariable logistic regression models were constructed, following the recommendation of Hosmer et al. [33], to identify which study variables were independently associated with the frequency and the number of days per week of PA among participants with diabetes and to assess possible changes from year 2014 to 2020. Adjusted odds ratios (ORs) with 95% confidence intervals (95% CIs) are provided as the measure of association of the multivariable models.
The statistical software used was STATA 14.0 (Stata Statistical Software Version 14. StataCorp LP, College Station, TX, USA).

Sensitivity Analysis
To assess if diabetes was associated with the dependent variables, and if this association can be explained by other sociodemographic or clinical covariates besides age and gender, we conducted a multivariable analysis using logistic regression with the entire study population and including all those covariates that showed a significant relationship with the PA.

Ethical Aspects
Any investigator can freely download the databases of the EHISS2014 and the EHISS2020 from the Spanish Ministry of Health website [14]. According to Spanish legislation, for investigations conducted with public access anonymous data provided by the health authorities, the approval of an ethics committee is waived. Figure S1 is the flowchart of participant's selection. Before matching, people with diabetes were significantly older than those without this condition in both surveys, 68.61 ± 13.40 vs. 51.71 ± 17.85 in the EHISS2014 (p < 0.001) and 70.25 ± 12.83 vs. 53.84 ± 18.07 (p < 0.001) in the EHISS2020. Regarding gender, the proportion of females was higher among those without diabetes than among those with diabetes (54.1% vs. 51.9%; p = 0.045 in the in the EHISS2014 and 53.3% vs. 49.6% in the EHISS2020; p < 0.001).

Shown in
The number of participants aged 18 years or older with self-reported physicians diagnosed diabetes were 1945 in the EHISS2014 and 2150 in the EHISS2020. After excluding those with missing data, the total number of matched couples reached 1852 and 1889 in these surveys, respectively.
The distribution of people with diabetes according to study variables for the two EHISS is shown in Table 1. The proportion of females decreased significantly from 52.5% in 2014 to 48.8% in 2020 (p = 0.023), whereas the mean age increased from 68.2 to 69.7 years (p < 0.001). The educational level improved overtime. Regarding clinical variables, people with diabetes had better self-rated health in year 2020 when compared to 2014 and a significant reduction was found for self-reported COPD, mental disease, and body mass index. On the other hand, alcohol consumption rose significantly from 37.3% to 41.2% (p = 0.014).
The proportion of people with diabetes that had a medium or high frequency of PA improved from 48.3% in 2014 to 52.6% in 2020 (p = 0.009). A similar tendency was observed for the number of days per week of PA, with 68.5% in 2014 and 77.7% in 2020 being engaged in two or more days (p < 0.001).

Gender Differences in Self-Reported PA between Males and Females with Diabetes
Shown in Figure 1 are the frequency of PA and number of days per week of PA according to gender among people with self-reported diabetes included in the EHISS conducted in years 2014 and 2020. As can be seen in the figure, males with diabetes had a significantly higher frequency of medium or high PA (58.1% vs. 39.5% in 2014 and 74.0% vs. 63.5% in 2020) and two or more days per week of PA (59.0% vs. 45.8% in 2014 and 81.6% vs. 73.6% in 2020) than females with diabetes in both surveys (all p < 0.001). For both genders, a significant increment in the frequency of PA and number of days per week was observed from the EHISS2014 to the EHISS2020 (all p < 0.001). PA: physical activity. COPD: chronic obstructive pulmonary disease. p value for differences between EHISS 2014 and EHISS 2020. p value was obtained using the chi-square test.

Gender Differences in Self-Reported PA between Males and Females with Diabetes
Shown in Figure 1 are the frequency of PA and number of days per week of PA according to gender among people with self-reported diabetes included in the EHISS conducted in years 2014 and 2020. As can be seen in the figure, males with diabetes had a significantly higher frequency of medium or high PA (58.1% vs. 39.5% in 2014 and 74.0% vs. 63.5% in 2020) and two or more days per week of PA (59.0% vs. 45.8% in 2014 and 81.6% vs. 73.6% in 2020) than females with diabetes in both surveys (all p < 0.001). For both genders, a significant increment in the frequency of PA and number of days per week was observed from the EHISS2014 to the EHISS2020 (all p < 0.001).

Differences in the Self-Reported PA between Participants with Diabetes and Age-Gender-Matched Non-Diabetic Subjects
Once the participants were matched, and both surveys joined, the medium or high frequency of PA was found in 59.4% of participants without diabetes and 50.5% of those with diabetes (p < 0.001). The frequency of PA was significantly higher among controls without diabetes than among cases with diabetes when the analysis was stratified by any of the sociodemographic variables shown in Table 2.
The proportion of people with diabetes that reported doing PA two or more days per week was significantly lower than for those without diabetes (73.1% vs. 78.3%; p < 0.001). As found for the frequency of PA, controls had higher number of days per week than cases after stratification by any sociodemographic variables.

Differences in the Self-Reported PA between Participants with Diabetes and Age-Gender-Matched Non-Diabetic Subjects
Once the participants were matched, and both surveys joined, the medium or high frequency of PA was found in 59.4% of participants without diabetes and 50.5% of those with diabetes (p < 0.001). The frequency of PA was significantly higher among controls without diabetes than among cases with diabetes when the analysis was stratified by any of the sociodemographic variables shown in Table 2.
The proportion of people with diabetes that reported doing PA two or more days per week was significantly lower than for those without diabetes (73.1% vs. 78.3%; p < 0.001). As found for the frequency of PA, controls had higher number of days per week than cases after stratification by any sociodemographic variables.
For both sub-populations (cases and controls), being a male, younger age, higher educational level, and living with a partner were associated to higher frequency and number of days of PA.
The frequency of PA and the number of days of PA according to clinical variables and lifestyles among participants with diabetes and matched controls without diabetes is shown in Table 3.
For most categories of the variables shown in Table 3, the proportions of matched controls without diabetes who reported medium or high frequency of PA or practicing two or more days per week were significantly higher than among those with diabetes.
Among people with diabetes, "Very good/good" self-rated health and not reporting any of the chronic conditions analyzed were associated to higher frequency and number of days of PA. Regarding lifestyles, participants with no diabetes who had alcohol consumption and lower BMI reported more PA than those with diabetes.

Multivariable Analysis to Determine which Study Variables Were Associated with Reporting PA among People with Diabetes
The results of the multivariable logistic regression model to identify, among participants with diabetes, which variables were independently associated with the frequency and the number of days per week of PA are shown in Table 4.
After adjusting for all the variables shown in the table, being a male was significantly associated to reporting a medium or high PA (OR 1.52; 95% CI 1.31-1.75) and to engaging in PA two or more day per week (OR 1.20; 95% CI 1.02-1.41). All age groups under 75 years reported more PA than the elderly.
The results of the multivariable model evidenced that having a higher educational level was associated to more frequency and greater number of days per week of PA. Furthermore, "Very good/good" self-rated health was also associated to more PA. On the other hand, self-reported COPD, heart diseases, stroke, mental disorders, active smoking, or a BMI over 30 were variables associated to lower frequency and number of days of PA.
After adjusting for possible confounders, the proportion of participants with diabetes who reported medium or high PA increased by 18% (OR 1.18; 95% CI 1.03-1.35) from 2014 to 2020. Furthermore, the increment in those who engaged in ≥2 days per week of PA was 62% (OR 1.62; 95% CI 1.39-1.88).

Sensitivity Analysis
Shown in Table S2 are the results of the multivariable analysis to assess if diabetes was associated with the frequency of PA and the number of days of PA, after controlling for all the sociodemographic or clinical covariates. The results found, with the entire study population, are very similar to those reported for participants with diabetes with any of the two dependent variables used. Therefore, being a male, younger age, higher educational level, good self-rated health, not suffering from concomitant chronic conditions or not being obese, and not consuming tobacco were factors associated to more PA.
Finally, the sensitivity analysis confirmed that people with diabetes reported a medium or high PA (OR 0.87; 95% CI 0.86-0.96) and being engaged in ≥2 days per week of PA (OR 0.91; 95% CI 0.82-0.99) significantly less than participants without diabetes.

Discussion
Our work showed an increase in the prevalence of self-reported PA from 2014 to 2020 in Spanish adults with diabetes, as well as a better perceived self-rated health. Studies of PA trends in the population with diabetes are limited [27], but until now, they have mostly shown stable or unfavorable trends [15,27,28]. Jimenez et al. found for Spaniards with diabetes aged older than 65 years using data from health surveys from 1995 to 2006 a greater proportion of a sedentary lifestyle over time [15]. Zhao et al. in the United Sates observed a stable trend in PA frequency in adult with diabetes from 1996 to 2005 [28].
The improvement observed in our study population is significant, but it is still insufficient: in 47% of persons with diabetes, the frequency of PA was classified as sedentary or low, and 22.3% dedicated one or no days per week to engage in any PA. Other Spanish studies obtained similar discouraging prevalence of PA in adults with diabetes, in all cases below the recommended levels [10,16,17,25,26]. Sarria Santamera et al., using the 2017 Spanish National Health Survey (SNHS), reported 36.3% physical inactivity among 1496 adults with diabetes [16]. López-Sánchez et al. estimated a prevalence of physical inactivity, measured through the International Physical Activity Questionnaire, of 35.4% in his diabetes population of 1014 persons in year 2020 [17]. These data are possibly even worse if we consider that previous studies have shown that people with diabetes frequently overestimate their levels of PA [34,35]. In a systematic review analyzing adherence to PA in individuals with type 2 diabetes, results ranged from 32% to 100%, with a median of 58%, although in only one study, PA compliance was the primary outcome [36].
As expected, subjects with diabetes engaged less in PA than controls matched for age and gender, and this finding was confirmed in the sensitivity analysis. This lower PA has been reported in a sample of over 100,000 adults in Germany using data from national population health surveys between 1997 and 2018, with lower prevalence of PA among people with obesity and diabetes than among people with normal weight and no diabetes [29]. Similar results have been found in the US population in the years 2016 and 2017, where 44.2% of 4860 persons over 65 years of age with diabetes or prediabetes reported PA two or three times a week compared to 48.1% from a matched sample without diabetes [30]. A recent systematic review of the literature found that these results are independent of the measurement instrument or study location [27].
We also found marked gender differences in the practice of PA in subjects with diabetes in favor of males. These differences have been previously observed in Spanish and international studies [16,25,26,31]. In Spain, a multicenter population study on adherence to healthy lifestyles in type 2 diabetes patients found that male gender was the variable most strongly associated with adequate compliance with nutrition and PA recommendations [26]. A systematic review and meta-analysis on gender differences in PA in type 2 diabetes adults throughout life concluded that these do not occur among adolescents but do appear with a remarkable magnitude in the older population [31].
The identification of females as a target group in which to increase PA adherence is especially relevant if we consider that females with diabetes have higher risk of suffering coronary disease than males with diabetes, with greater sequelae and mortality [37]. This excess of cardiovascular risk has been associated with poorer control of risk factors in females, mainly in the prediabetic phase, associated with a higher percentage of body fat, and together with other sociodemographic variables [38].
In addition to being a female, we found that higher age, lower educational level, and current smoking were associated to lower PA in the multivariable analysis. These associations have been confirmed in studies conducted among people with diabetes and in the general population [25,26,29,30,36].
Older people with diabetes are especially vulnerable, since they have a greater risk of suffering the consequences of inactivity, such as frailty, sarcopenia, and other chronic diseases, compared to younger ones [39]. Yang et al. in a recently published work on participation in PA among American elderly adults with type 2 diabetes found that beyond sociodemographic variables, personal factors such as extroversion and low neuroticism in adherence to exercise were factors that should be considered to optimize the results of health improvement strategies based on lifestyle modifications [40].
The presence of comorbidities such as COPD, heart disease and stroke, mental disorders, and obesity was also associated with less PA in our population with diabetes. These findings are like those reported in the literature [16,30,41], and possibly have a two-way direction causality. Furthermore, related to this association with comorbidities, and as expected, better self-perceived health status was found as an independent predictor of more PA, agreeing with other studies [42].
The association between obesity and less PA among people with diabetes is especially relevant due to the benefit in glycemic control and the cardiovascular risk profile of weight loss that can be achieved with PA [2,9,29]. Intervention approaches in this subgroup of patients should be individualized, multidisciplinary, and always considering their specific barriers to PA [2,9,29,43].
More efforts are needed to promote greater adherence to PA recommendations in the Spanish population with diabetes. It has been estimated that the improvement in PA in these patients can be associated with global savings in direct and indirect costs that represent 35% of the total healthcare expenditure in Spain [16].
Health policies should promote PA at the population level, through campaigns that publicize its benefits, the recommended levels of PA, and that "every movement count", especially if it is combined with a reduction in sitting position [39]. Physical education should be encouraged from the school environment, as well as promote the creation and improve accessibility to spaces for the practice of exercise [39]. Efforts directed at the general population will result in an improvement in the population with diabetes, but it is important to implement lifestyle improvement programs specifically aimed at the most vulnerable groups of people with diabetes, such as females, the elderly, and obese individuals, to help them to initiate and maintain the benefits of an active lifestyle. Linking exercise to leisure and socialization can be a particularly positive strategy in these patients [2,[42][43][44][45].
Health professionals should be aware of this patient profile and include in their ongoing training plans knowledge about PA, benefits, safety considerations, and practical management options [44]. Any informative action that is addressed to these groups from health centers will have a positive impact. The recommendation of exercise must be individualized and centered on the patient with diabetes, and a comprehensive and multidisciplinary approach, which can be optimized by including graduates in physical activity and sports sciences in therapeutic teams. Mobile health interventions (mHealth) can represent an alternative or complement to face-to-face programs, although their results still require further effectiveness studies [42][43][44][45][46].

Limitations
Our study has limitations that should be mentioned. First, the causality direction cannot be addressed due to the study design. Second, the questions used for self-reported PA and diabetes have not been validated in the EHISS. However, a Spanish study showed a specificity of >95% and a sensitivity > 70% for self-reported diabetes using medical records as the gold standard [47]. In epidemiological research, the use of self-reported PA and diabetes within population surveys has been previously reported [15][16][17]24,27,28,30,31,35,36,41,42]. Third, the EHISS lacks specific information on diabetes, such as type, complications, duration of the disease, and treatments. Forth, as for any interview survey, the existence of recall errors or socially desirable responses must be considered. Fifth, another relevant limitation of our investigation is that only two levels of frequency of PA (up to 1 day/week and >2 days/week) were available to the participants, so we cannot assess a possible dose-response relationship. Furthermore, as commented before, previous studies have reported that overestimation of PA is possible when it is self-reported [34,35]. However, our intention using this question was to identify those individuals that had a very severe degree of sedentarism, because they did not even walk for at least 10 min continuously in a week more than once. Sixth, important factors such as the patients' area of living (rural vs. urban) are not collected by the EHISS, so they could not be analyzed. Previous works have found that this factor might be a factor linked to PA trends [48].
Finally, the response rates for the EHISS 2014 were 61% and for the EHISS2020 it was 59%; therefore, a non-response bias could have affected our results [21,22]. As commented before, due to the COVID-19 pandemic, the collection method during the last months of the EHISS2020 was modified and the effect of this change or of the pandemic itself on our results cannot be ruled out [13].

Conclusions
In conclusion, the trend of adherence to self-reported PA in Spanish adults with diabetes is favorable but insufficient. The prevalence of PA in this diabetes population is low and does not reach the levels of PA in the general population. The main factors associated with lower adherence to exercise were female gender, being older, lower educational level, worse self-rated health, the presence of comorbidities, obesity, and smoking. Our result could help to improve the design and implementation of public health strategies to improve PA among people with diabetes. However, it is necessary to deepen the research with studies specifically designed to identify attitudes and barriers to PA in people with diabetes, as well as the role of new technologies to improve adherence to PA in this population.

Supplementary Materials:
The following supporting information can be downloaded at: https://ww w.mdpi.com/article/10.3390/jcm12062443/s1. Figure S1: Flowchart of participant's selection. Table S1: Definition of variables according to the questions included in the European Health Interview Surveys in Spain conducted in years 2014 and 2020; Table S2: Sensitivity analysis. Variables associated with medium or high physical activity and with number of days per week of physical activity ≥ 2 days in the study population. Results of multivariable logistic regression analysis.