Incorporating Informatics for Integrating Biology and the Bedside (i2b2) into Predoctoral Trainee Curriculum to Evaluate Student‐Generated Hypotheses

Abstract

As part of the Clinical and Translational Science Institute predoctoral TL1 training program at the Pennsylvania State University, a multidisciplinary team of predoctoral trainees representing the Chemistry, Neurosurgery, Nutritional Sciences, and Public Health Sciences departments were introduced to the NIH‐sponsored Informatics for Integrating Biology and the Bedside (i2b2) database to test the following student‐generated hypothesis: children with iron deficiency anemia (IDA) are at increased risk of attention deficit‐hyperactivity disorder (ADHD). Children aged 4–12 and 4–17 years were categorized into IDA and control groups. De‐identified medical records from the Penn State Milton S. Hershey Medical Center (HMC) and the Virginia Commonwealth University Medical Center (VCUMC) were used for the analysis. Overall, ADHD prevalence at each institution was lower than 2011 state estimates. There was a significant association between IDA and ADHD in the 4–17‐year‐old age group for all children (OR: 1.902 [95% CI: 1.363–2.656]), Caucasian children (OR: 1.802 [95% CI: 1.133–2.864]), and African American children (OR: 1.865 [95% CI: 1.152–3.021]). Clinical and Translational Science Award (CTSA) infrastructure is particularly useful for trainees to answer de novo scientific questions with minimal additional training and technical expertise. Moreover, projects can be expanded by collaborating within the CTSA network.

Introduction

The National Center for Advancing Translational Sciences (NCATS) was established by the National Institutes of Health (NIH) in 2012 to accelerate the delivery of new treatments to improve health. The establishment of NCATS was preceded by the Clinical and Translational Science Award (CTSA), which created a national consortium of more than 60 institutions to promote clinical and translational research and encourage collaboration.1 The Pennsylvania State University Clinical and Translational Science Institute (CTSI) TL1 training program provides predoctoral students from diverse research backgrounds with unique opportunities to acquire the experience and knowledge required to become successful independent clinical and translational scientists. As part of the curriculum, TL1 trainees attend a weekly multidisciplinary colloquium that studies translational research seminars at two campuses (the Pennsylvania State University main campus in University Park, PA and the College of Medicine in Hershey, PA). One seminar (“Developmental Origins of Altered Stress Resilience: Impact of Iron Deficiency” given by Dr. Rodney Johnson of the University of Illinois, Urbana‐Champaign2, 3) piqued the interest of a cohort of TL1 scholars representing the departments of Chemistry, Neurosurgery, Nutritional Sciences, and Public Health Sciences. They formed a research team, identified a scientific problem, and developed a hypothesis that could be tested using the Informatics for Integrating Biology and the Bedside (i2b2) database.

i2b2 is a NIH‐sponsored National Center for Biomedical Computing based at Partners HealthCare Systems in Boston, MA. It was developed to accelerate translational research by making genomic and clinical data accessible to researchers, and to promote more personalized healthcare through the use of new prognostic and diagnostic tools.4, 5 This database allows researchers with a range of bioinformatics skills to study electronic medical records without requiring IRB approval. At Penn State Milton S. Hershey Medical Center (HMC), i2b2 contains de‐identified electronic medical records starting from 2011 and is updated monthly. HMC is a 777‐bed facility, with 128 of those located in the Children's Hospital. It had 29,140 admissions and 960,831 outpatient visits in the 2013–2014 fiscal year.6 A similar database at a partner CTSA institution, Virginia Commonwealth University Medical Center (VCUMC) in Richmond, VA, provided additional data from their catchment region from 2007 to present. VCUMC is a 1,125‐bed facility with an additional children's hospital (Children's Hospital of Richmond) that contains 168 beds. In the 2012–2013 fiscal year, VCUMC had 34,105 admissions and 580,005 outpatient visits.7 Differences in race and age also distinguish these two catchment pools. Richmond, VA is racially diverse, with 40.8% Caucasians and 50.6% African Americans whereas Hershey, PA is primarily Caucasian (83.5%), with only 6.2% African Americans. With respect to age demographics, only 18.6% of the Hershey population was less than 18 years old compared to 29.1% of the Richmond population.8

Structured discussion helped students refine their hypothesis from a general inquiry about the role of iron in cognitive development to a specific question that could be directly queried using the i2b2 database. Further discussion and review of the literature base helped students recognize that a retrospective analysis could be used to test a related hypothesis, that iron deficiency anemia (IDA) and attention‐deficit/hyperactivity disorder (ADHD) are positively associated in children.

IDA consists of two concomitant diagnoses: iron deficiency and anemia. Iron deficiency is diagnosed by the presence of two or more abnormal laboratory measurements, typically serum ferritin, serum iron, and soluble transferrin receptor. Anemia is identified by low hemoglobin and hematocrit, resulting in reduced oxygen transport. Those at greatest risk for IDA are premature infants, infants fed only breast milk or formula not fortified with iron, those with reduced iron intake or poor absorption from the diet, and those with blood loss from surgery or other pathologies.9 Iron acts as a cofactor for neurotransmitter metabolism, DNA repair, and DNA synthesis.10, 11, 12 During neurodevelopment, iron is also involved in functions such as myelination and dendritic arborization.13, 14, 15, 16, 17, 18 Thus, IDA could contribute to impaired brain development, learning, and memory.2, 13, 19 Iron is also critical for regulating dopaminergic activity and iron deficiency has been implicated in neurobehavioral disorders such as ADHD.20, 21, 22, 23, 24, 25, 26

ADHD is a complex behavioral disorder, which results in inattention and/or hyperactivity and impulsivity.27 It is typically diagnosed in school‐aged children as the school environment is often particularly challenging for individuals with ADHD symptoms. The pathophysiology of ADHD is not completely understood but several studies have indicated that iron deficiency may be related to the development of ADHD. Compared to controls, children with ADHD have been shown to have significantly reduced serum ferritin.28 Multivariate logistic regression has also demonstrated that low serum iron (OR: 2.81 [95% CI: 1.72–4.53]) and low ferritin (OR: 2.53 [95% CI: 1.81–3.45]) were among the significant predictors of ADHD in a Qatari population of children 5–18 years old.26 There was also increased risk of ADHD (OR: 1.67 [95% CI: 1.29–2.17]) in an iron deficient Taiwanese population of children <18 years old.22 Iron supplementation in ADHD children has shown some effectiveness, although larger studies are needed to determine efficacy, dose, and the most appropriate method of administration.25, 29, 30, 31, 32, 33 Nonetheless, the relationship between iron deficiency and ADHD remains controversial as not all studies have found an association between the two conditions.34, 35 Therefore, we used i2b2 to test whether children in the HMC and VCUMC databases with IDA were more likely to have a concurrent diagnosis of ADHD compared to controls.

Methods

Data were obtained from de‐identified electronic medical records using the i2b2 database. Students were trained to use i2b2 to determine the prevalence of ADHD in a control and IDA population of school‐aged children. Individuals were identified by ICD‐9 diagnosis codes for ADHD (314.01) and iron deficiency anemia (280.1, 280.8, and 280.9). Data were further stratified into two age groups: 4–12 and 4–17 years old. An additional stratification by race was included to compare the African American and Caucasian populations. Queries were designated in i2b2 by the following: ADHD and IDA (Group 1 = 314.01, Group 2 = 280.1, 280.8, 280.9, Group 3 = age stratification, Group 4 = race stratification); IDA only (Group 1 = 280.1, 280.8, 280.9, Group 2 = age stratification, Group 3 = race stratification); Control and ADHD (Group 1 = 314.01, Group 2 = age stratification, Group 3 = gender stratification, Group 4 = exclude 280.1, 280.8, 280.9); and Control (Group 1 = age stratification, Group 2 = race stratification, Group 3 = exclude 280.1, 280.8, 280.9 and 314.01). Counts were obtained for each group to determine the prevalence of disease in each population. The same query terms were used for each institution.

ADHD prevalence between HMC and VCUMC were analyzed using Fisher's exact test (two‐sided). Comparison of ADHD prevalence between the IDA and control populations was performed using Fisher's exact test (two‐sided) and risk ratio via a 2 × 2 contingency table. Analysis was performed using SPSS (Version 22; IBM, Armonk, NY, USA).

Results

The prevalence of ADHD in our HMC and VCUMC populations was low compared to the 2011 estimated prevalence for both states36 (Table 1). There was no significant difference (p = 0.741) in ADHD prevalence in the control populations aged 4–17 years at HMC (3.41%) and VCUMC (3.37%). Overall, we found that children with IDA are at higher risk of ADHD. The association was significant for both age groups at both institutions, except for patients aged 4–12 years at HMC, although this group trended towards significance (p = 0.093). The strongest association was seen when including all children aged 4–17 years from HMC and VCUMC in the analysis (OR: 1.902 [95% CI: 1.363–2.656], p < 0.0001; Table 2). The association between IDA and ADHD was also significant for the combined HMC and VCUMC 4–12‐year‐old age group (OR: 1.959 [95% CI: 1.284–2.987], p = 0.005).

Table 1.

Comparison of 2011 ADHD prevalence by state and 2015 ADHD prevalence by institution

Institution	ADHD prevalence in controls from this study aged 4–17 years	2011 state prevalence—currently diagnosed ADHD from CDC (17)
Pennsylvania State University Milton S. Hershey Medical Center	3.41%	9.3%
Virginia Commonwealth University	3.37%	9.1%

Table 2.

ADHD prevalence in children with iron deficiency anemia compared to controls

Institution	Prevalence of ADHD in IDA population aged 4–12 years	Prevalence of ADHD in control population aged 4–12 years	OR (95% CI)	p Value aged 4–12 years*	Prevalence of ADHD in IDA population aged 4–17 years	Prevalence of ADHD in control population aged 4–17 years	OR (95% CI)	p Value aged 4–17 years*
Pennsylvania State University Milton S. Hershey Medical Center	8/140 (5.71%)	1,807/56,306 (3.21%)	1.828 (0.894–3.737)	0.093	16/230 (6.95%)	3,082/90,496 (3.41%)	2.121 (1.274–3.529)	0.009
Virginia Commonwealth University Medical Center	15/270 (5.55%)	1,571/58,398 (2.69%)	2.128 (1.261–3.590)	0.008	21/361 (5.82%)	3,032/89,771 (3.37%)	1.767 (1.135–2.750)	0.018
All	23/410 (5.61%)	3,378/114,704 (2.94%)	1.959 (1.284–2.987)	0.005	37/591 (6.26%)	6,114/180,267 (3.39%)	1.902 (1.363–2.656)	<0.0001

Comparison of ADHD prevalence in control populations between institutions was performed by Chi squared test (p < 0.0001 for ages 4–12 years and p = 0.741 for ages 4–17 years).

*Fisher's exact test used for obtaining 2‐sided p value.

When stratifying by race (Caucasians and African Americans), the association between ADHD and IDA was significant in both the 4–12‐year‐old (OR: 2.251 [95% CI: 1.096–4.623], p = 0.033) and 4–17‐year‐old (OR: 2.089 [95% CI: 1.188–3.673], p = 0.015) Caucasian populations of HMC. No significant association was found in the Caucasian VCUMC population. However, analysis of the combined 4–17‐year‐old Caucasian populations from HMC and VCUMC did demonstrate significantly increased risk (OR: 1.802 [95% CI: 1.133–2.864], p = 0.019) for developing ADHD (Table 3).

Table 3.

ADHD prevalence in Caucasian children with iron deficiency anemia compared to controls

Institution	Prevalence of ADHD in IDA population aged 4–12 years	Prevalence of ADHD in control population aged 4–12 years	OR (95% CI)	p Value aged 4–12 years*	Prevalence of ADHD in IDA population aged 4–17 years	Prevalence of ADHD in control population aged 4–17 years	OR (95% CI)	p Value aged 4–17 years*
Pennsylvania State University Milton S. Hershey Medical Center	8/116 (6.89%)	1,607/50,434 (3.18%)	2.251 (1.096–4.623)	0.033	13/192 (6.77%)	2,737/81,459 (3.36%)	2.089 (1.188–3.673)	0.015
Virginia Commonwealth University Medical Center	2/118 (1.69%)	577/30,141 (1.91%)	0.883 (0.218–3.583)	1.000	6/157 (3.82%)	1,252/47,356 (2.64%)	1.463 (0.646–3.315)	0.315
All	10/234 (4.27%)	2,184/80,575 (2.71%)	1.602 (0.849–3.023)	0.153	19/349 (5.44%)	3,989/128,815 (3.10%)	1.802 (1.133–2.864)	0.019

Comparison of ADHD prevalence in control populations between institutions was performed by Chi squared test (p < 0.0001 for both ages 4–12 years and ages 4–17 years).

*Fisher's exact test used for obtaining 2‐sided p value.

For African American children, the association between IDA and ADHD was significant for both age groups in the VCUMC population only. No significant association was found in the HMC children. When combining African American children from both institutions, increased risk was seen in both the 4–12 (OR: 2.200 [95% CI: 1.247–3.882], p = 0.011) and 4–17 (OR: 1.865 [95% CI: 1.152–3.021], p = 0.015) age groups (Table 4).

Table 4.

ADHD prevalence in African American children with iron deficiency anemia compared to controls

Institution	Prevalence of ADHD in IDA population aged 4–12 years	Prevalence of ADHD in control population aged 4–12 years	OR (95% CI)	p Value aged 4–12 years*	Prevalence of ADHD in IDA population aged 4–17 years	Prevalence of ADHD in control population aged 4–17 years	OR (95% CI)	p Value aged 4–17 years*
Pennsylvania State University Milton S. Hershey Medical Center	0/24 (0%)	200/5,872 (3.40%)	N/A	N/A	3/38 (7.89%)	345/9,037 (3.82%)	2.160 (0.661–7.056)	0.177
Virginia Commonwealth University Medical Center	13/152 (8.56%)	994/28,257 (3.52%)	2.565 (1.448–4.545)	0.003	15/204 (7.35%)	1,780/42,415 (4.19%)	1.812 (1.069–3.072)	0.034
All	13/176 (7.39%)	1,194/34,129 (3.50%)	2.200 (1.247–3.882)	0.011	18/242 (7.44%)	2,125/51,452 (4.13%)	1.865 (1.152–3.021)	0.015

Comparison of ADHD prevalence in control populations between institutions was performed by Chi squared test (p = 0.694 for ages 4–12 years and p = 0.104 for ages 4–17 years).

*Fisher's exact test used for obtaining 2‐sided p value.

Discussion

Our retrospective study is an example of the unique educational experiences offered by a CTSI TL1 training program. The multidisciplinary, discussion‐based format of the TL1 colloquium encouraged this type of discovery education and the introduction to contemporary tools, such as i2b2. Exposure to medical informatics databases in the early stages of graduate training provides students with the tools and experience needed to design and test preliminary clinical and/or translational science hypotheses. Currently, the HMC i2b2 database includes over 600,000 patients. This provides clinical and translational science researchers access to de‐identified medical records from a large population without IRB approval. Forming collaborations with other CTSA institutions can further increase the size of the population available for study and may improve demographic distributions. For example, Hershey, PA consists of predominantly Caucasian population (83.5% Caucasian, 6.2% African American), whereas Richmond, VA is 50.6% African American and 40.8% Caucasian8; therefore, expanding our study to include VCUMC increased the diversity of our study population. Using these databases also introduces concepts of data inclusion and exclusion as it requires students to properly identify their population of interest by excluding patients with confounding diagnoses, lab values, or medications.

Our study indicated that children with IDA are approximately twice as likely to be concomitantly diagnosed with ADHD. This association was identified by comparing the difference in ADHD prevalence in iron deficient anemic and control populations of children at two institutions (HMC and VCUMC). Interestingly, we found that this association differed in certain populations when the analysis was stratified by race and age. Including additional CTSA institutions and thus studying more children throughout the United States, as well as accounting for additional medical variables would be of interest to confirm the results found in the HMC and VCUMC populations.

The prevalence of ADHD in our population was much lower than previously reported. Visser et al. estimated state and nationwide ADHD prevalence for children aged 4–17 years in 2011 based on parent‐report in the National Survey of Children's Health. The prevalence of children diagnosed with ADHD was 9.3% in Pennsylvania and 9.1% in Virginia.36 The National Health Interview Survey, an ongoing cross‐sectional health survey in the United States, has reported similar rates of ADHD prevalence for Caucasian and African American populations. As of December 2013, ADHD prevalence among children aged 3–17 years was 9.9% in Caucasian children and 8.8% in African American children.37 The lower ADHD prevalence in our population compared to previously reported estimates36, 37 may be due to limitations in i2b2 or differences in how the data was obtained. For example, a diagnosis of ADHD may be made by individuals outside of the hospital medical systems (e.g., social workers, counselors, therapists, psychiatrists, and neurologists). In this case, the diagnosis would not be captured in i2b2 and the individual would not be categorized as having ADHD in our analysis.

i2b2 is a beneficial tool for testing preliminary hypotheses; however, when training students, its limitations should be noted to ensure that the data is properly analyzed. For instance, because ICD‐9 diagnosis codes are obtained through the institution's billing system, the data are prone to human error and only include diagnoses that were billed through the medical center. Additionally, since the HMC and VCUMC databases date back to 2011 and 2007, respectively, if a patient was diagnosed with ADHD prior this and was not billed for the diagnosis since then, they would not be included in the ADHD group produced by the i2b2 query. While this is unlikely, it may partially account for the lower ADHD prevalence in our study population compared to reported state and national prevalence rates. Furthermore, i2b2 currently lacks the ability to include confounding variables such as smoking, diet, and genetic influences; therefore, results should be interpreted with some caution. Additional limitations may also be presented by specific i2b2 queries. For instance, for our study, an array of exclusion criteria could have been included to limit the analysis only to healthy children without confounding diagnoses. However, the children analyzed by Visser et al. were included based on random digit‐dialing and not all were necessarily healthy.36 Since this was the study to which we compared our prevalence data, we chose to forego adding additional exclusion criteria to our queries.

This unique educational experience incorporated the use of an electronic medical record database into the CTSI training program and allowed TL1 trainees to develop and test a scientifically relevant hypothesis. While i2b2 was ideal for our purposes, other databases, (e.g., the National Health and Nutrition Examination Survey [NHANES]; Surveillance, Epidemiology, and End Results [SEER]; and Behavioral Risk Factor Surveillance System [BRFSS]) could be utilized in a similar fashion to test student‐generated hypotheses. By using these databases, trainees can expand their understanding of the type of data available and its limitations, how to best utilize the data to address the hypothesis, and how to develop future studies.

Funding

This publication was supported, in part, by Grant UL1 TR000127 and TL1 TR000125 from the National Center for Advancing Translational Sciences (NCATS).