Characterization of Primary IGF-1 Deficiency in a Cohort of Canadian Children with Short Stature Using a Novel Algorithm Tailored to Electronic Medical Records

(1) Background: Severe primary insulin-like growth factor-I deficiency (SPIGFD) is a rare disorder causing short stature in children due to low insulin-like growth factor 1 (IGF-1) levels. Given the sparsity of reported cases of SPIGFD worldwide, the condition may be underdiagnosed, potentially preventing affected children from receiving therapy with recombinant human IGF-1 (rhIGF-1). Our objective was to determine the prevalence of SPIGFD among children with short stature at a large pediatric tertiary care center through the use of a novel electronic medical record (EMR) algorithm. (2) Methods: We queried our EMR using an algorithm that detected all children seen at our center between 1 November 2013 and 31 August 2021 with short stature and low IGF-1. We then conducted chart reviews, applying established diagnostic criteria for those identified with potential SPIGFD. (3) Results: From a cohort of 4863 children with short stature, our algorithm identified 30 (0.6%) patients with potential SPIGFD. Using chart reviews, we determined that none of these patients had SPIGFD. (4) Conclusions: Our algorithm can be used in other EMRs to identify which patients are likely to have SPIGFD and thus benefit from treatment with rhIGF-1. This model can be replicated for other rare diseases.


Introduction
Growth hormone (GH) is essential for growth in children.Deficiencies or defects in the function of GH at, for example, receptor or signaling levels may result in short stature [1].Insulin-like growth factor 1 (IGF-1) is a protein that mediates the action of GH.IGF-1 deficiency can disrupt the effects of GH and drastically impede muscle, bone, and overall growth [2].
There are two main forms of IGF-1 deficiency: primary and secondary.Primary IGF-1 deficiency is characterized by inadequate production of IGF-1 despite sufficient GH [3], while secondary IGF-1 deficiency includes other causes of low IGF-1, such as hypothyroidism, malnutrition, or growth hormone deficiency [2].Severe primary IGF-1 deficiency (SPIGFD) is characterized by low basal IGF-1 levels, short stature, growth hormone sufficiency, and the exclusion of secondary forms of IGF-1 deficiency.SPIGFD results primarily from GH receptor mutations, post-receptor signaling defects, or inactive IGF-1 [3].Clinical features of IGF-1 deficiency include short stature, low bone mineral density, and obesity or reduced lean body mass [1,4].
Children 2024, 11, 727 2 of 12 Since the recognition of IGF-1 deficiency as a primary endocrine disorder in the early 1990s, genetic mutations that cause low or inactive IGF-1 have been identified, such as mutations in the GH receptor gene in those with Laron syndrome [4].These discoveries led to the development of treatments that bypass the mutated GH receptors, such as recombinant human IGF-1 (rhIGF-1) [2,5].To date, research related to IGF-1 and SPIGFD has focused on the effectiveness, safety, and long-term effects of rhIGF-1 treatments [1,[5][6][7], while details regarding its diagnosis and the prevalence of this ultra-rare disease are sparse.
The diagnosis of SPIGFD remains challenging, as secondary causes of IGF-1 deficiency exist [8].Given the sparsity of reported cases of SPIGFD worldwide, we suspect that the condition is underdiagnosed, likely due to low IGF-1 levels being attributed to secondary causes.Reports of the prevalence of SPIGFD vary [9,10], and to our knowledge, the prevalence in Canada has not been previously reported.While rhIGF-1 is safe and effective in improving both linear growth rate and final adult height in children with SPIGFD [5,7,11], an electronic medical record (EMR) algorithm that could identify missed cases of SGPIFD does not exist.As a result, it is difficult to identify children that may benefit from rhIGF-1, potentially preventing affected children from receiving adequate therapy.The aim of our study was to establish a diagnostic algorithm using an electronic medical record system (EMR) and to determine the prevalence of SPIGFD among children with short stature at a large Canadian pediatric tertiary care center.

Study Design
We leveraged our EMR, Epic (Epic Systems, Verona, WI, USA), to identify our cohort and automate as much of the chart reviews as possible.To accomplish this, we developed an algorithm to identify patients with potential SPIGFD at the Children's Hospital of Eastern Ontario (CHEO), Ottawa, ON, Canada.We obtained approval from the Research Ethics Board at CHEO.The requirement for informed consent was waived given the retrospective nature of the study and the fact that all the data collected were de-identified.

Study Population
We assessed the medical records of all children between the ages of 0 and 18 years with an outpatient encounter at CHEO between 1 November 2013 and 31 August 2021.Our inclusion criteria were based on diagnostic criteria for SPIGFD: short stature, defined as a height of <3.0 standard deviations (SD) below the mean for age and sex using the World Health Organization (WHO) growth charts for Canada [12], and low IGF-1, defined as basal levels < 2.5th percentile for sex, age, and/or pubertal status and/or bone age [13].At our institution, we used two different IGF-1 assays: the IDS-iSYS assay (Immunodiagnostic Systems; Tyne & Wear, United Kingdom) from November 2013 to September 2017 and the Liaison assay (DiaSorin; Vercelli, Italy) as of October 2017.We defined IGF-1 values according to the appropriate assay and adjusted for pubertal stage and/or bone age as needed (depending on which one was available in the medical chart in cases where there was more than one low IGF-1 result based on pubertal stage or bone age, we used the first recorded instance of low).We defined pubertal status according to Tanner staging [14] and estimated bone age using the Greulich and Pyle atlas as reported by the reading pediatric radiologist [15].

Identification of Children with SPIGFD
The Epic algorithm we developed aimed to identify children meeting criteria for SPIGFD and exclude those with secondary IGF-1 deficiency, resulting in a limited number of manual chart reviews necessary to identify children with SPIGFD.
Our algorithm included six sequential steps (Figure 1).First, we identified all children with short stature (height ≤ −3.0 SD) based on the WHO growth charts for Canada [12] (step 1).Of those, we found the number of children with at least one IGF-1 value < 2.5th percentile for sex, age, and/or pubertal status and/or bone age (step 2).In cases where there was more than one low IGF-1 result, we used the first recorded instance of low IGF-1.In step 3, we identified the number of children with short stature recorded within one year of a low IGF-1 value, excluding those who either did not have an IGF-1 < 2.5th percentile or did not have a height recorded within 1 year of a low IGF-1.In step 4, we excluded children with secondary causes of IGF-1 deficiency, such as malnutrition, hypopituitarism, hypothyroidism, chronic treatment with supraphysiologic doses of steroids, liver disease, or other potential secondary causes for low IGF-1 values.We defined secondary causes of IGF-1 deficiency using the International Classification of Diseases 10th Revision (ICD-10) codes [16] for specific diagnoses (malnutrition, hypothyroidism, GH deficiency, and diseases of the liver) and body mass index (BMI) of <5th percentile on the WHO growth charts for Canada [12], suggestive of malnutrition.These diagnoses were taken from either the visit diagnoses or the problem lists; both EMR elements are frequently updated, and clinicians are unable to close encounters without indicating a visit diagnosis.A complete list of the ICD-10 codes used to define secondary causes of IGF-1 deficiency is available in Appendix A. We categorized the remaining children identified into six categories of abnormal growth, based on those defined by Teissier et al.: (1) transient pubertal slowing of growth with resumption of normal growth velocity at pubertal onset; (2) endocrine diseases (i.e., constitutional delay of growth, septo-optic dysplasia, etc.); (3) small for gestational age (as defined by length and/or weight at birth ≤ −2 SD); (4) genetic disorders (i.e., abnormal gene test results or multiple birth defects); ( 5) idiopathic short stature (ISS, as defined by height ≤ −2 SD with no identified etiology); and ( 6) adverse effects of medications on growth [9].

Statistical Analysis
We used descriptive statistics to describe the proportion of children with potential SPIGFD identified by our Epic algorithm among children with short stature (height ≤ −3.0 SD) and low IGF-1 assessed at CHEO within the study period.All other clinical characteristics were similarly described using descriptive statistics.Clinical characteristics of patients identified with potential SPIGFD were verified via manual chart review to ensure we were only including children who met our criteria (i.e., we manually excluded children with GH deficiency, and secondary forms of IGF-1 deficiency not excluded by the algorithm; (step 5)).Growth hormone deficiency was defined as a suboptimal peak GH response to stimulation by clonidine and/or arginine.At our institution, we used two different analyzers with different peak GH definitions: peak value < 5.4 ng/mL using Dxl analyzers (Beckman Coulter; Brea, CA, USA) from 2005 to 27 November 2019, and peak value < 7.4 ng/mL using Roche analyzers (Hoffmann-La Roche, Basel, Switzerland) from 27 November 2019 onwards.Finally, we conducted an indepth chart review, with 2 clinical experts (CB and MER) assessing the remaining patients for growth and IGF-1 trajectories to exclude any child who had a good height response to GH treatment (defined as an absolute height gain >0.5 SD within a 12-month period [17]), normalization of IGF-1 levels on GH treatment, or spontaneous normalization of IGF-1 and/or growth (step 6).

Statistical Analysis
We used descriptive statistics to describe the proportion of children with potential SPIGFD identified by our Epic algorithm among children with short stature (height ≤ −3.0 SD) and low IGF-1 assessed at CHEO within the study period.All other clinical characteristics were similarly described using descriptive statistics.

Study Cohort
Figure 1 shows the results of our automated Epic algorithm and chart review.All patient encounters during our study window were screened, which yielded a total of 4863 unique children meeting our definition of short stature (step 1).Of the cohort, 233 (4.8%) had low IGF-1 (step 2), and 124 (2.5%) had both short stature and low IGF-1 within one year of each other (step 3).

Description of Patients with Potential SPIGFD
Our Epic algorithm identified 30 children (0.6% of all children with short stature; 24% of all children with short stature and IGF-1 deficiency) with potential SPIGFD (step 4; Appendix A).The identified children (47% female) had a mean (SD) age of 7.0 (5.0) years, and mean (SD) height and weight z-scores of −3.5 (0.5) and −2.0 (1.5) at the time of their first low IGF-1 result.GH stimulation tests were normal in 47% of the cohort, and were not performed in the remaining 53%.The mean (SD) follow-up duration for these 30 children was 5.3 (1.7) years, and the cohort had a mean (SD) change in height SDS of 0.6 (2.0) over this time.The classification of these children is displayed in Figure 2.
Of the 30 children identified by the algorithm with potential SPIGFD, we excluded 19 who were found to have secondary IGF-1 deficiency via the manual chart review (step 5).We assessed growth and IGF-1 trajectories in the remaining 11 children with potential SPIGFD, and all 11 patients were excluded: 5 children because of appropriate GH response, 4 due to spontaneous normalization of IGF-1, and 2 based on spontaneous normalization of both growth and IGF-1 (step 6) (Table 1).The median time to spontaneous IGF-1 normalization was 10.8 months (range 6.1-24.1 months).Ultimately, upon manual review, there were no children who met the criteria for a clinical diagnosis of SPIGFD during our study window.mean (SD) height and weight z-scores of −3.5 (0.5) and −2.0 (1.5) at the time of their first low IGF-1 result.GH stimulation tests were normal in 47% of the cohort, and were not performed in the remaining 53%.The mean (SD) follow-up duration for these 30 children was 5.3 (1.7) years, and the cohort had a mean (SD) change in height SDS of 0.6 (2.0) over this time.The classification of these children is displayed in Figure 2. Of the 30 children identified by the algorithm with potential SPIGFD, we excluded 19 who were found to have secondary IGF-1 deficiency via the manual chart review (step 5).We assessed growth and IGF-1 trajectories in the remaining 11 children with potential SPIGFD, and all 11 patients were excluded: 5 children because of appropriate GH response, 4 due to spontaneous normalization of IGF-1, and 2 based on spontaneous normalization of both growth and IGF-1 (step 6) (Table 1).The median time to spontaneous One child had adverse effects of medications (chronic glucocorticoid use), 1 child had transient prepubertal slowing of growth, 2 children were small for gestational age, 3 children had idiopathic short stature, 8 children had endocrine diseases (6 with constitutional delay of growth and 2 with septic optic dysplasia), and 13 children had genetic disorders; 1 child had both an endocrine disease (constitutional delay of growth) and a genetic disorder (Duchenne Muscular Dystrophy), and 1 child had both an endocrine disease (constitutional delay of growth) and spontaneous normalization of growth/IGF-1.

Discussion
In this study, by leveraging our EMR, we created an electronic algorithm for characterizing the prevalence of SPIGFD.Through this algorithm, combined with minimal manual chart reviews, we determined that there were no children meeting criteria for a clinical diagnosis of SPIGFD among children with short stature at our large Canadian pediatric tertiary care center.While the algorithm identified 30 children with potential SPIGFD, representing a prevalence of 0.6% among children with short stature, we ultimately excluded all 30 patients upon manual chart review, resulting in a prevalence of 0%.
Out of 4863 children with short stature assessed, none met the diagnostic criteria for SPIGFD.Previous studies in France, England, and the United States have reported SPIGFD prevalence rates ranging from 1.2 to 25% using varied definitions and source populations [9,10,18,19].For example, using a similar definition to that used in the present study (height < −3 SDS, serum IGF-1 levels < 2.5th percentile, GH sufficiency, and absence of causes of secondary IGF-1 deficiency), Teissier and colleagues found the prevalence of SPIGFD to be 0.2% in a large prospective cohort of French children with short stature [9].Conversely, Edouard and colleagues found a SPIGFD prevalence of 20% in a small sample of prepubertal children with short stature, using a definition of height SDS ≤ −2.5, IGF-1 SDS ≤ −2, and GH sufficiency [10].
There may be several reasons that account for our finding of no children with SPIGFD relative to previous studies.First, we followed our cohort for an average of over 5 years, whereas previous studies followed cohorts for 6 months [9] or for an unspecified duration [10,18,19].The importance of a longer duration of follow-up/documentation of the persistence of low IGF-1 is highlighted by the fact that the minimum time to spontaneous normalization of IGF-1 seen in our cohort was >6 months (median 10.8 months, range 6.1-24.1 months).While a single low IGF-1 value is insufficient to diagnose SPIGFD, it is expected that IGF-1 values should never normalize without rhIGF-1 therapy in the context of true SPIGFD.There is no consensus on the number of repeat assessments that should be conducted or over what time frame.There are also no internationally recognized standards regarding IGF-1 assay methodologies or sample collection and storage [20].However, IGF-1 generation tests may be helpful in the diagnosis of SPIGFD [21].
It is unlikely that we may have missed cases of SPIGFD, as our institution is the only referral center for pediatric patients within our catchment area, and assessment of IGF-1 levels is part of our routine pediatric endocrinology workup for children with short stature.It is possible that some children meeting criteria for SPIGFD may have been seen in a clinic where IGF-1 testing is not necessarily routine, such as the genetics clinic.However, it is standard practice to refer children with short stature to endocrinology, where an IGF-1 is routinely performed.While some practice guidelines propose algorithms for when to test IGF-1 and/or treat with rhIGF-1 (e.g., [22], there is significant practice variation in this realm.Implementation of clear clinical algorithms guiding diagnosis and management would help to ensure all eligible children are referred and tested appropriately in the context of short stature. Third, unlike previous studies, which only assessed prepubertal children [9,10], we included all children regardless of pubertal status.It is possible that some children who initially presented as having potential SPIGFD may have normalized their IGF-1 during puberty, as was the case in our study (normalization of IGF-1 during puberty is not possible in cases of SPIGFD).
Our study has demonstrated that data stored in an EMR can successfully be leveraged to identify patients with rare diseases that are not easily classified using ICD codes.Data from EMRs have been successfully used to identify children with other complex conditions such as post-traumatic stress disorder [23], attention deficit hyperactivity disorders with psychiatric comorbidities [24], and adults with rare conditions such as allergic bronchopulmonary aspergillosis [25].Automated algorithms and EMR data mining have been identified as potential ways to identify patients with rare diseases [26].Leveraging clinical data to identify such patients has several advantages, including a reduction in the number of manual chart reviews required and the potential transferability to other centers using EMR.Another benefit is the potential to identify patients for relevant therapies or interventions, as would be the case with SPIGFD and the available treatment with rhIGF1.To our knowledge, this is the first time that an automated algorithm has been developed and used to identify children with SPIGFD.Leveraging our EMR resulted in fewer manual chart abstractions, which minimized reviewer time, and ensured standardization in the application of the diagnostic criteria throughout the algorithm steps, reducing potential bias.This proof-of-concept design will enable our methodology to be expanded to other pediatric tertiary care centers using a similar EMR.It will also allow for the potential for data from many centers to be easily pooled (with relatively minimal resource requirements), ultimately resulting in a larger population from which a definitive prevalence of SPIGFD could be determined.A large, multi-center sample may also provide an opportunity to utilize machine learning to develop a real-time clinical practice algorithm to identify which patients are likely to have SPIGFD and thus benefit from treatment with rhIGF-1.
Our study has several strengths.To our knowledge, this is the largest cohort of children and the longest duration of follow-up included in a SPIFD prevalence study.Also, the problem lists and diagnoses in our EMR are frequently updated by clinicians and are therefore highly accurate, as clinicians are unable to close a clinical encounter or visit encounter without specifying a visit diagnosis.The results of our study should also be interpreted, given some potential limitations.Given the retrospective nature of the study, we were limited by the data available in health records, which may have resulted in some cases of SPIGFD being missed (for example, if an IGF-1 level was not completed in a patient with short stature).However, we cast a wide net with the intention of maximizing the sensitivity of our algorithm (by interpreting IGF-1 levels based on puberty and bone age rather than age only) in order to minimize the premature or incorrect exclusion of patients.Due to the ultrarare prevalence of SPIGFD, in order to maximize our chances of identifying a patient meeting these criteria, we utilized as wide of a date range as possible (i.e., from the earliest date EMR data were available until the time the work was completed).This left no room for a validation cohort, and therefore the algorithm could not be appropriately evaluated, which should be the subject of future work.

Conclusions
From a cohort of 4863 children with short stature, we identified no children with clinical SPIGFD.Though divergent from other prevalence studies, our findings are robust given our large sample size, long follow-up duration, and comprehensive chart reviews.The design, methodology, and results of this study will help inform future studies to identify potential cases of SPIGFD and ultimately help with the identification of children who may benefit from targeted therapy.A future study using a large, multi-center sample may also leverage our findings to utilize machine learning to develop a real-time clinical practice algorithm to identify which patients are likely to have SPIGFD and thus benefit from treatment with rhIGF-1.This model can also be replicated for other rare diseases.

13 Figure 1 .
Figure 1.Flow chart of the algorithm to identify cases of SGPIFD via the Epic database, starting with all children seen at CHEO in our study time frame and ending with those meeting the diagnosis criteria for SPIGFD.

Figure 1 .
Figure 1.Flow chart of the algorithm to identify cases of SGPIFD via the Epic database, starting with all children seen at CHEO in our study time frame and ending with those meeting the diagnosis criteria for SPIGFD.

Figure 2 .
Figure 2. Categorization of children (n = 30) with potential SPIGFD identified by the Epic algorithm.One child had adverse effects of medications (chronic glucocorticoid use), 1 child had transient prepubertal slowing of growth, 2 children were small for gestational age, 3 children had idiopathic short stature, 8 children had endocrine diseases (6 with constitutional delay of growth and 2 with septic optic dysplasia), and 13 children had genetic disorders; 1 child had both an endocrine disease (constitutional delay of growth) and a genetic disorder (Duchenne Muscular Dystrophy), and 1 child had both an endocrine disease (constitutional delay of growth) and spontaneous normalization of growth/IGF-1.

Figure 2 .
Figure 2. Categorization of children (n = 30) with potential SPIGFD identified by the Epic algorithm.One child had adverse effects of medications (chronic glucocorticoid use), 1 child had transient prepubertal slowing of growth, 2 children were small for gestational age, 3 children had idiopathic short stature, 8 children had endocrine diseases (6 with constitutional delay of growth and 2 with septic optic dysplasia), and 13 children had genetic disorders; 1 child had both an endocrine disease (constitutional delay of growth) and a genetic disorder (Duchenne Muscular Dystrophy), and 1 child had both an endocrine disease (constitutional delay of growth) and spontaneous normalization of growth/IGF-1.

Table 1 .
Patients identified with potential SPIGFD after manual chart review for exclusion of secondary IGF-1 deficiency.Patients were assessed for GH response and spontaneous normalization of growth and IGF-1 levels (n = 11) (Step 5).