Abstract
Purpose:
Recurrent pathogenic copy number variants (pCNVs) have large-effect impacts on brain function and represent important etiologies of neurodevelopmental psychiatric disorders (NPDs), including autism and schizophrenia. Patterns of health care utilization in adults with pCNVs have gone largely unstudied and are likely to differ in significant ways from those of children.
Methods:
We compared the prevalence of NPDs and electronic health record-based medical conditions in 928 adults with 26 pCNVs to a demographically-matched cohort of pCNV-negative controls from >135,000 patient-participants in Geisinger’s MyCode Community Health Initiative. We also evaluated 3 quantitative health care utilization measures (outpatient, inpatient, and emergency department (ED) visits) in both groups.
Results:
Adults with pCNVs (24.9%) were more likely than controls (16.0%) to have a documented NPD. They had significantly higher rates of several chronic diseases, including diabetes (29.3% in participants with pCNVs vs. 20.4% in participants without pCNVs) and dementia (2.2% in participants with pCNVs vs. 1.0% in participants without pCNVs), and twice as many annual emergency department visits.
Conclusion:
These findings highlight the potential for genetic information – specifically, pCNVs - to inform the study of health care outcomes and utilization in adults. If, as our findings suggest, adults with pCNVs have poorer health and require disproportionate health care resources, early genetic diagnosis paired with patient-centered interventions may help to anticipate problems, improve outcomes, and reduce the associated economic burden.
Introduction
Rare, pathogenic copy number variants (pCNVs) of large effect size represent important genetic etiologies of autism spectrum disorder (ASD), intellectual disability (ID), schizophrenia (SCZ), epilepsy, and other neurodevelopmental psychiatric disorders (NPDs)1–3. Dozens of NPD-associated pCNVs are now known, collectively contributing to a significant subset of clinical brain disorders, with shared underpinnings in known biochemical and neurobiological pathways. As such, the study of rare pCNVs has important relevance for all NPD research and may ultimately advance precision medicine efforts toward the prevention and treatment of mental health disorders.
We previously identified pCNVs in 0.8% of our large health care cohort3, consistent with reports from other unselected population-based studies, such as the UK Biobank (1%),4,5 Estonian Genome Center (0.7%),6 and deCODE Genetics in Iceland (1.16%).7 Diagnostic yields for pCNVs in specific NPD studies range from 3% in SCZ to >15% in ID, with higher percentages among subgroups with significant cognitive impairments and/or congenital anomalies.1,2,8,9 In clinical settings, chromosomal microarray analysis has long been recommended for the detection of pCNVs in children with global developmental delay/ID and ASD,10 with some expert groups also endorsing its use in individuals with SCZ.9,11,12 In the United States, detection of pCNVs is increasingly incorporated into exome sequencing analysis, which is now also recommended as a first-tier diagnostic test in children with ID and ASD.13,14 Similar consensus guidelines do not exist for adults with NPDs, and the vast majority of those with pCNVs have not had clinical genetic testing.3 As rare, NPD-related copy number variants are often highly penetrant, many of these adults live with chronic medical, cognitive, psychiatric, and neurological disorders without ever knowing the underlying genetic etiologies that unify these seemingly unrelated medical conditions.
A compelling argument in favor of establishing a genetic diagnosis in pediatric populations relates to the avoidance of unnecessary diagnostic testing and the initiation of pre-emptive monitoring for known medical comorbidities.15 For example, pCNVs are frequently associated with congenital anomalies, and establishing an etiological diagnosis informs treatment approaches and long-term medical surveillance. By contrast, patterns of health care utilization in adults with pCNVs have gone largely unstudied and are likely to differ in significant ways from those of children with genetic disorders. The 22q11.2 deletion syndrome, with its variable and complex somatic phenotypes and 25% risk for SCZ, is often cited to illustrate the medical relevance of pCNVs in adults.16 There are also other examples of pCNVs with adult-onset clinical manifestations that inform anticipatory guidance, such as maturity onset diabetes of the young (MODY5) in 17q12 deletions,17 morbid obesity in 16p11.2 deletions,18 and Birt-Hogg-Dube symptoms in 17p11.2 deletions.19
Apart from such descriptions of medical phenotypes in adults with specific rare disorders, few studies have examined the long-term health consequences of NPD-related pCNVs at a population level.3–5 In a previous study of Geisinger’s large health system cohort, we primarily explored the variable expression of NPD diagnoses among individuals with pCNVs while emphasizing the clinical and personal utility of disclosing genetic test results to adult research participants.3 We also highlighted the limitations of relying solely on electronic health record (EHR)-based diagnostic codes for NPD phenotyping. In 2019, Crawford et al4 conducted a large-scale investigation of UK Biobank participants and estimated the risk of developing common medical phenotypes among adults with 54 preselected pCNVs. The authors created a morbidity map on the basis of their phenome-wide study that exhaustively tested specific pCNVs (eg, 16p11.2 duplications) with a defined list of 58 curated diseases. The medical phenotypes found to be most strongly associated with pCNVs included renal failure, diabetes, hypertension, and obesity. The authors concluded that these increased medical risks among participants with pCNVs were unlikely to be related to the presence of NPDs, given the relative lack of developmental and psychiatric conditions in the UK Biobank cohort.
Significant gaps remain in our understanding of health care utilization across the lifespan of individuals with pCNVs beyond their selected ascertainment in pediatric genetics settings. Adults with pCNVs are long past the age when significant congenital anomalies would have been surgically addressed, although psychiatric concerns, including mood and psychotic disorders, typically emerge and require ongoing medical management in adulthood. The presence of a clinical psychiatric diagnosis, regardless of etiology, is known to be associated with increased morbidity, mortality, and medical costs,20,21 whereas the severity of mental illness also correlates with emergency department (ED) use.22 Recurrent ED users, 50% of whom in the United States have a mental health diagnosis,23 account for a disproportionate percentage of ED visits and expenditures. These and other predictors of health care utilization – defined here as the quantification of inpatient and outpatient medical resource use – can inform prevention and cost containment strategies. The recent availability of large-scale genomic data sets provides an important new resource for health care utilization research. Here, we leveraged genomic information from >135,000 patient-participants in Geisinger’s MyCode Community Health Initiative,24 along with linked EHR data, to evaluate medical conditions and health care utilization in adults with NPD-related pCNVs. We specifically focused on medical diagnoses from the Charlson Comorbidity Index,25 a well-established and standardized method of categorizing patient comorbidities in health care research, while also examining metrics related to inpatient, outpatient, and ED usage.
Materials and Methods
The DiscovEHR cohort
We analyzed exome sequencing results and linked EHR data from patient-participants in DiscovEHR, which is a subset of Geisinger’s MyCode Community Health Initiative and an ongoing research project in collaboration with the Regeneron Genetics Center.3 MyCode is a population-based health care cohort that consists of a predominantly rural adult population of European ancestry receiving care across the health system.24 For this analysis, we included data through November 14, 2019 on all DiscovEHR participants aged 18 years or older who had at least 1 EHR encounter at any Geisinger site in Pennsylvania.
Genomic and phenomic analyses
DNA sample preparation and exome sequencing for 135,883 DiscovEHR patient-participants were performed in collaboration with the Regeneron Genetics Center.24,26 Copy number calling of the exome sequence data was carried out on 135,848 cram files available at the time of study using the Copy number estimation using Lattice-Aligned Mixture ModelS algorithm.27 As part of post-CNV calling quality control (QC), only samples passing QC for CLAMMS were eligible for inclusion in the current study. We excluded samples with >40 CNV calls (n = 2520) or full or partial aneuploidies (n = 356) from further analysis. We evaluated the frequency of 31 recurrent, segmental duplication-mediated pCNVs in the DiscovEHR cohort, as previously described.3 In brief, we conservatively restricted selection to large recurrent pCNVs having ClinGen Dosage Sensitivity scores of 3 during the analysis period (September 2019-March 2021). For this study, we combined overlapping subregions of 1q21.1 and 22q11.2 into single pCNVs instead of considering each region separately, resulting in a total of 26 CNVs (Table 1).
Table 1.
Pathogenic CNVs (n=26) among eligible study participantsa
| Copy Number Variant | Dosage | Count | GRCh38 Coordinates |
|---|---|---|---|
|
| |||
| 1q21.1 (GJA5) | Deletion | 64 | chr1:147.11–147.92 |
| 1q21.1 (GJA5) | Duplication | 113 | chr1:147.11–147.92 |
| 3q29 (DLG1) | Deletion | 6 | chr3:196.03–197.62 |
| 7q11.23 (ELN) | Deletion | 5 | chr7:73.33–74.73 |
| 7q11.23 (ELN) | Duplication | 13 | chr7:73.33–74.73 |
| 8p23.1 (GATA4) | Deletion | 1 | chr8:8.26–11.91 |
| 10q23 (BMPR1A) | Deletion | 1 | chr10:79.92–86.98 |
| 15q11.2q13 BP1–3 (UBE3A) | Deletion | 3 | chr15:22.78–28.14 |
| 15q11.2q13 BP1–3 (UBE3A) | Duplication | 5 | chr15:22.78–28.14 |
| 15q13.3 BP4–5 (CHRNA7) | Deletion | 75 | chr15:30.84–32.15 |
| 15q24 (SIN3A) | Deletion | 2 | chr15:72.67–75.68 |
| 16p11.2 distal (SH2B1) | Deletion | 31 | chr16:28.81–29.04 |
| 16p11.2 (TBX6) | Deletion | 71 | chr16:29.64–30.19 |
| 16p11.2 (TBX6) | Duplication | 108 | chr16:29.64–30.19 |
| 16p13.11 (MYH11) | Deletion | 86 | chr16:15.42–16.20 |
| 17p12 (PMP22) | Deletion | 47 | chr17:14.19–15.52 |
| 17p12 (PMP22) | Duplication | 56 | chr17:14.19–15.52 |
| 17p11.2 (RAI1) | Deletion | 3 | chr17:16.91–20.30 |
| 17q11.2 (NF1) | Deletion | 3 | chr17:30.78–31.94 |
| 17q11.2 (NF1) | Duplication | 6 | chr17:30.78–31.94 |
| 17q12 (HNF1B) | Deletion | 12 | chr17:36.46–37.85 |
| 17q12 (HNF1B) | Duplication | 65 | chr17:36.46–37.85 |
| 22q11.2 (TBX1) | Deletion | 17 | chr22:18.92–20.29 |
| 22q11.2 (TBX1) | Duplication | 134 | chr22:18.92–20.29 |
| 22q11.2 distal | Deletion | 2 | chr22:21.44–23.31 |
| 22q11.2 distal | Duplication | 1 | chr22:21.44–23.31 |
Two adults with multiple pCNVs are included in the table (one with 1q21.1 duplication / 17p12 deletion and one with 1q21.1 deletion / 22q11.2 duplication). Multiple pCNVs are included in the counts.
EHR extraction used structured demographic data, encounter details, and International Classification of Diseases (ICD)-9 and -1028 codes generated from routine clinical care. We defined an NPD phenotype to include diagnoses of attention deficit hyperactivity disorder, ASD, bipolar and related disorder, cerebral palsy, communication disorder, epilepsy, ID, motor disorder, obsessive compulsive and related disorder, other neurodevelopmental disorder, SCZ spectrum and other psychotic disorder, and specific learning disorder based on the presence of a relevant ICD-9 or ICD-10 code (for details, see definitions of NPD ICD codes in Martin et al).3 The conditions included in the Charlson Comorbidity Index25 were used to describe the health of the cohort.
Statistical analysis
Adults with pCNVs were matched to DiscovEHR participants without any of the 26 recurrent pCNVs described earlier. Matching was performed with a ratio of 1:5 using the following variables: age (±2 years), sex, race/ethnicity, insurance type, and duration in the EHR (± 0.5 years). Descriptive statistics are presented as means, SDs, medians, and inter-quartile ranges for continuous variables and frequency and percentages for categorical variables. Demographic variables and comorbidities were compared between the groups using random effects models and the clustered Wilcoxon rank sum test29 to account for clustering due to matching. For conditions that were of low frequency (≤2%) in both groups, the Fisher’s exact test was used. Any variable found to significantly differ between groups was considered a potential confounding variable. To estimate the effect of having a pCNV on health care utilization, the negative binomial distribution was used to model the number of outpatient, ED, and inpatient visit counts, controlling for the matching and potential confounding variables. A series of models were fit to assess how confounding affected the estimate. First, a model was fit with only the matching variables (model 1). Next, body mass index (BMI) was added (model 2) because it is known that some pCNVs are related to obesity.30 Then, any phenotypes found to significantly vary between groups were added to the model (model 3). The next model added NPDs (model 4). Finally, we tested for an interaction between pCNV and NPD presence. To account for varying lengths in the EHR, an offset term−the natural logarithm of the EHR duration−was included in the models to allow for a standardized interpretation as the average number of visits per year. Results were presented as the adjusted average number of visits and rate ratios (RRs). Corresponding 95% CI were also calculated. Analyses were performed using SAS v9.431 and R v4.0.332
Results
After application of QC and study inclusion criteria, 928 and 125,919 individuals with and without pCNVs remained in our study, respectively. The 928 individuals with pCNVs represent 26 recurrent pathogenic deletions and duplications (Table 1). Of 873 individuals, 94% were successfully matched to 5 controls. The remaining eligible participants (n = 55) were matched using relaxed criteria for age (±5 years). Of these, 53 of the 55 were matched to 5 controls, 1 participant was matched to 4 controls, and another was matched to 1 control, resulting in a total of 4635 matched controls. On average, the groups were 51 years old, 62% female, and had approximately 13 years of data in the EHR (Table 2). Study participants with a pCNV had a significantly higher percentage of chronic conditions, including congestive heart disease, peripheral vascular disease, dementia, chronic pulmonary disease, diabetes, and renal disease. They also had an increased average BMI compared with their matched controls (33.3 kg/m2 vs 31.5 kg/m2, P < 0.0001). Because of the low number of participants sharing specific pCNVs relative to the larger aggregate pCNV group and the overall DiscovEHR cohort, individual pCNV subcohorts were underpowered for meaningful multivariate statistical analysis.
Table 2.
Demographics and clinical conditionsa in adults with and without pCNVs
| pCNV Positive (N=928) | pCNV Negative (N=4635) | p-value | |
|---|---|---|---|
| Follow-up duration (years) | 12.8 (7.2, 17.5) | 12.9 (7.2, 17.6) | N/Ab |
| Age | 51.0 (16.56) | 51.0 (16.54) | N/Ab |
| Sex (Male) | 355 (38.2%) | 1771 (38.2%) | N/Ab |
| Body Mass Index (BMI) | 33.3 (9.85) | 31.5 (8.18) | <0.001c |
| White | 898 (96.8%) | 4490 (96.9%) | 0.982 |
| Hispanic | 12 (1.3%) | 55 (1.2%) | 0.977 |
| Insurance | |||
| Geisinger Health Plan | 436 (47.0%) | 2180 (47.0%) | 0.998 |
| Medicaid/Medicare | 177 (19.1%) | 880 (19.0%) | 0.993 |
| Other insurance | 315 (33.9%) | 1575 (34.0%) | 0.999 |
| Deceased (Yes) | 54 (5.8%) | 237 (5.1%) | 0.351 |
| Smoker (Yes) | 489 (52.7%) | 2553 (55.1%) | 0.173 |
| Neurodevelopmental Psychiatric Disorders (NPD) (Yes) | 231 (24.9%) | 741 (16.0%) | <0.001c |
| Myocardial Infarction | 37 (4.0%) | 202 (4.4%) | 0.604 |
| Congestive Heart Disease | 92 (9.9%) | 337 (7.3%) | 0.004c |
| Peripheral Vascular Disease | 78 (8.4%) | 292 (6.3%) | 0.014c |
| Cerebrovascular Disease | 80 (8.6%) | 384 (8.3%) | 0.722 |
| Dementia | 20 (2.2%) | 48 (1.0%) | 0.001c |
| Chronic Pulmonary Disease | 276 (29.7%) | 1215 (26.2%) | 0.025c |
| Rheumatic Disease | 32 (3.4%) | 204 (4.4%) | 0.180 |
| Peptic Ulcer Disease | 15 (1.6%) | 91 (2.0%) | 0.5659 |
| Mild Liver Disease | 66 (7.1%) | 360 (7.8%) | 0.492 |
| Diabetes without Chronic Complicationd | 272 (29.3%) | 944 (20.4%) | <0.001c |
| Diabetes with Chronic Complicationd | 12 (1.3%) | 34 (0.7%) | 0.052 |
| Diabetes | 273 (29.4%) | 944 (20.4%) | <0.001c |
| Hemiplegia or Paraplegia | 8 (0.9%) | 48 (1.0%) | 0.582 |
| Renal Disease | 133 (14.3%) | 514 (11.1%) | 0.002c |
| Neoplasm | 105 (11.3%) | 609 (13.1%) | 0.112 |
| Moderate or Severe Liver Disease | 4 (0.4%) | 17 (0.4%) | 0.768 |
| Metastatic Solid Tumor | 10 (1.1%) | 77 (1.7%) | 0.2448 |
| Human Immunodeficiency Virus (HIV)/ Acquired Immunodeficiency Syndrome (AIDS) | 0 (0.0%) | 10 (0.2%) | 0.386 |
| Congenital malformations of the nervous system | 1 (0.1%) | 6 (0.1%) | >0.999 |
| Congenital malformations of the heart and great vessels | 11 (1.2%) | 18 (0.4%) | 0.001c |
| Congenital malformations of the urinary system | 4 (0.4%) | 9 (0.2%) | 0.251 |
| Cleft lip and palate | 1 (0.1%) | 0 (0.0%) | 0.167 |
| Congenital malformations of the genital organs | 1 (0.1%) | 3 (0.06%) | 0.518 |
| Migraine | 43 (4.6%) | 167 (3.6%) | 0.120 |
Conditions included are those present in the Charlson Comorbidity Index
Matching variable, comparison was not performed
p-value <0.05
Although analyzed separately here, conditions were combined into a single diabetes variable in the utilization models
Table 3 shows the descriptive statistics of the utilization outcomes. The median number of years in the EHR was 12.8 and 12.9 in the pCNV and non-pCNV groups, respectively. The pCNV cohort had greater utilization across all visit types after controlling for the matching variables. Specifically, reported in Table 3, those with a pCNV had, on average, 6% more outpatient visits (RR = 1.06, 95% CI = 1.01–1.11), 35% more ED visits (RR = 1.35, 95% CI = 1.20–1.52), and 24% more inpatient visits (RR = 1.24, 95% CI= 1.10–1.39) over this 13-year period compared with the matched non-pCNV group.
Table 3.
Cumulative Number of Healthcare Encounters Stratified by pCNV Statusa
| pCNV (N=928) | Non-pCNV (N=4635) | p-valueb | |
|---|---|---|---|
| Follow-up duration in years | 12.8 (7.2, 17.5) | 12.9 (7.2, 17.6) | N/Ac |
| Total number of OP visits | 52.0 (23.8, 93.0) | 48.0 (24.0, 86.0) | 0.0220 |
| Total number of ED visits | 2.0 (0.0, 6.0) | 1.0 (0.0, 4.0) | <0.0001 |
| Total number of IP visits | 1.0 (0.0, 2.0) | 0.0 (0.0, 2.0) | 0.0004 |
Shown as median (interquartile range)
Controls for the matching variables (age, sex, race/ethnicity, insurance type, EHR duration)
Matching variable, comparison was not performed
To aid in the interpretation of the RR values, the estimated yearly average number of visits is also reported in Table 4 (model 1). As described before, the models were extended to control for BMI (model 2). ED and inpatient visits remained significantly increased in the pCNV group; however, the effect of having a pCNV was no longer significant for outpatient visits. After further controlling for medical conditions that were found to vary significantly between groups (model 3), the pCNV effect was attenuated but remained significant for both ED and inpatient visits. For ED, those with a pCNV had a yearly average of 0.51 visits compared to 0.41 in the non-pCNV group (RR = 1.26, 95% CI = 1.12–1.41), representing a 26% increase. Similarly, for inpatient visits, the pCNV group had an estimated yearly average of 0.18 visits compared to 0.14 in the non-pCNV group (RR = 1.14, 95% CI = 1.02–1.27). Next, the presence of an NPD was included in the models (model 4). After controlling for NPDs, having a pCNV was only significant for ED, with those participants having 20% more ED visits (0.56 vs 0.46, RR = 1.20, 95% CI = 1.07–1.35) per year.
Table 4.
Adjusted average number of visits per year based on modelinga
| pCNV+ group (N=928) | pCNV− group (N=4635) | RR (95% CI) | p-value | ||
|---|---|---|---|---|---|
|
| |||||
| Outpatient Visits | Total | 65,600 | 298,078 | ||
| Model 1 | 5.98 (5.45, 6.56) | 5.66 (5.20, 6.15) | 1.06 (1.01, 1.11) | 0.0220 | |
| Model 2 | 5.91 (5.39, 6.48) | 5.65 (5.20, 6.14) | 1.05 (0.998, 1.098) | 0.0629 | |
| Model 3 | 5.49 (5.02, 6.00) | 5.48 (5.06, 5.94) | 1.00 (0.96, 1.05) | 0.9664 | |
| Model 4 | 5.75 (5.26, 6.28) | 5.83 (5.38, 6.32) | 0.99 (0.94, 1.03) | 0.5539 | |
|
| |||||
| ED Visits | Total | 4,659 | 16,873 | ||
| Model 1 | 0.55 (0.44, 0.69) | 0.41 (0.33, 0.50) | 1.35 (1.20, 1.52) | <0.0001 | |
| Model 2 | 0.55 (0.44, 0.69) | 0.41 (0.34, 0.50) | 1.35 (1.20, 1.52) | <0.0001 | |
| Model 3 | 0.51 (0.41, 0.64) | 0.41 (0.34, 0.50) | 1.26(1.12, 1.41) | 0.0001 | |
| Model 4 | 0.56 (0.45, 0.69) | 0.46 (0.38, 0.56) | 1.20 (1.07, 1.35) | 0.0014 | |
|
| |||||
| Inpatient Visits | Total | 1,745 | 6,698 | ||
| Model 1 | 0.21 (0.17, 0.26) | 0.17 (0.14, 0.20) | 1.24 (1.10, 1.39) | 0.0004 | |
| Model 2 | 0.21 (0.17, 0.26) | 0.17 (0.14, 0.20) | 1.25 (1.11, 1.40) | 0.0002 | |
| Model 3 | 0.18 (0.14, 0.22) | 0.16 (0.13, 0.19) | 1.14 (1.02, 1.27) | 0.0229 | |
| Model 4 | 0.19 (0.15, 0.23) | 0.18 (0.15, 0.22) | 1.06 (0.95, 1.18) | 0.3041 | |
Models defined below and shown as average visits (95% confidence intervals):
- Model 1: Matching variables only
- Model 2: Model 1 + BMI
- Model 3: Model 2 + congestive heart disease, peripheral vascular disease, dementia, chronic pulmonary disease, diabetes, renal disease
- Model 4: Model 3 + NPD
Finally, we tested for an interaction between NPDs and pCNVs. The interaction was not significant in the models for outpatient visits (P = .4100) and inpatient visits (P = .9817). However, the interaction was found to be significant in the ED model (P = .0389). To help with the interpretation of this interaction, we compared the effect of having a pCNV among those with and without NPDs (Figure 1). Among those with an NPD, there was a nonsignificant effect of having a pCNV (RR = 0.97, 95% CI = 0.77–1.22, P = .8237). By contrast, among those without NPDs, those with a pCNV had 28% more yearly ED visits than those without a pCNV (RR = 1.28, 95% CI = 1.13–1.46, P = .0002).
Figure 1.
Adjusted average number of ED visits per year, stratified by pCNV and NPD status
Discussion
In this retrospective study of Geisinger’s large, rural health care population, we compared the frequency of NPDs and EHR-based medical conditions from the Charlson Comorbidity Index in adults with 26 pCNVs to a cohort of demographically-matched controls without one of these pCNVs. We also evaluated 3 quantitative health care resource utilization measures (outpatient, inpatient, and ED visits) in both groups. Adults with pCNVs had higher rates of NPDs, chronic diseases, and resource utilization than matched controls, suggesting that genetic information may add value beyond that provided by clinical diagnoses alone for studying the determinants of health care utilization. One limitation of the current study is that, because of incomplete data, we did not directly examine the impact of income, occupation, education, or other parameters of socioeconomic status (SES). However, significant associations between SES and NPDs have been well described, and low SES is known to correlate with poorer health outcomes and higher ED utilization.33 Given the significantly elevated rate of NPDs in our pCNV cohort and the known strong association of NPDs with SES, our analyses in model 4 may account for some of the SES effect.
As expected, adults with pCNVs had significantly higher rates of NPDs than controls, with 25% of individuals having an ICD-9 or -10 code documenting a neurodevelopmental and/or psychiatric diagnosis. This finding very likely represents an underestimate of pCNV-related behavioral and cognitive symptoms. We have separately demonstrated the limitations of solely basing NPD estimates on ICD diagnoses in the EHR in a subset of adult DiscovEHR participants with pCNVs to whom genomic test results were disclosed as part of Geisinger’s Genomic Screening and Counseling program.3,34 In that ongoing study, 36 of 61 individuals (59%) with no evidence of NPD-related ICD-9/10 diagnostic codes were found on manual chart review to have 1 or more mental health diagnoses documented in unstructured EHR data elements, such as free-text physician notes; an additional 13 of 61 individuals (21%) described a relevant family and/or personal history of NPDs on interview. In total, 71% of adults with pCNVs thought to be unaffected through ICD-9/10 code analysis actually had evidence of NPDs on further investigation (Wain, personal communication).
The strong correlation of pCNVs with NPDs has important implications for identifying patient subgroups who might benefit from enhanced health care supports, as has been shown with the use of dedicated care “navigators,” specialized case managers who serve to coordinate health services for patients with cancer, diabetes, and other chronic conditions.35 Given the low accuracy of detecting at-risk NPD subgroups through automated EHR analysis, population-based diagnosis and disclosure of pCNVs, similar to existing protocols for the return of secondary genomic findings,36 may be a more effective strategy to develop an enhanced health care safety net for this vulnerable population.
With regard to chronic medical conditions, diagnostic codes for congestive heart failure, peripheral vascular disease, and chronic pulmonary disease in adults with pCNVs were all significantly higher than in controls, despite similar rates of tobacco use. The frequency of chronic renal disease and diabetes was also elevated in this population, consistent with previous findings in a broader study of pCNVs among adult participants in the UK BioBank.4 The increased prevalence of these chronic diseases may be related to significantly higher BMI among adults with pCNVs, potentially compounded by the well-known metabolic effects of some psychotropic medications for the treatment of NPDs. However, Owen et al5 found abnormal BMI and weight metrics associated with pCNVs in the UK BioBank, a population with relatively low rates of NPD diagnoses37 and, presumably, psychotropic medication use. Several pCNVs included in our current analysis are known to be associated with increased BMI, and it is possible that they are the primary drivers of the effect. Excess BMI is associated with significant morbidity and mortality, as well as health care utilization, even when BMI-specific survival is taken into consideration.38
Participants with pCNVs in this study were more than twice as likely as controls to have a diagnosis of dementia. This finding requires further exploration to determine whether dementia represents a previously undescribed extension of the known vulnerability conferred by NPD-related pCNVs to a continuum of brain disorders over the lifetime. Alternatively, the higher prevalence of dementia in the pCNV cohort could be directly linked to a secondary association with elevated rates of known dementia risk factors, such as diabetes, cardiovascular disease, and obesity.39
A small percentage of the increase in cardiovascular disease in this study could be explained by congenital heart defects, found in 1.2% of the pCNV cohort versus 0.4% of controls. The prevalence of other documented malformations in adults with pCNVs did not differ significantly from those without pCNVs. This relatively low rate of congenital anomalies in our adult cohort contrasts with their much higher reported prevalence in pediatric pCNV populations. This discrepancy likely reflects over-ascertainment of pCNVs in neonates with malformations and survivor bias in older adults with mild pCNV phenotypes. As such, efforts to assess the clinical utility of genetic testing in adults with NPDs need to look for associations beyond congenital anomalies, such as increased rates of common chronic diseases that are more likely to affect long-term health outcomes. Equally important, the nonstructural impacts of pCNVs, such as neurodevelopmental and psychiatric conditions that emerge during adolescence and adulthood, become a more significant driver of medical needs than congenital anomalies and may lead to different patterns of health care utilization in adults.
To broadly evaluate these patterns, we quantified inpatient, outpatient, and ED visits among adults with and without pCNVs. Total visits for all 3 encounter types were higher for individuals with pCNVs than for controls. The difference in outpatient visit frequency between the groups was not explained by demographic factors but could be accounted for by differences in BMI. The pCNV effect on inpatient health care utilization is not easily explained by demographic factors, BMI, or other medical phenotypes. Instead, it seems to be largely mediated by the association of the pCNVs with NPDs.
The largest pCNV impact on utilization was seen for ED visits, and the relationship remained significant across all models, suggesting that the effect of having a pCNV is not easily explained by demographic factors, BMI, other medical comorbidities, or NPDs. With regard to this outcome, the genetic information seems to offer something beyond what can be gleaned from clinical diagnoses based on ICD codes in the EHR. The significant interaction between NPDs and pCNV status suggests that something other than NPDs may be responsible for the increase in ED visits in the pCNV group. Alternatively, the brain disorders in the group with pCNVs could be qualitatively different from those in matched controls (eg, increased severity or a different distribution of NPD diagnoses), given the known limitations of ICD codes for capturing NPD phenotypes. As a group, pCNVs could be associated with disorders and/or NPD symptoms that are more likely to escape detection by ICD codes in the EHR.
EDs are a high-cost health care setting and a focus of efforts to reduce avoidable utilization, and further exploration of this finding is important to identify pCNV-related behaviors that might be amenable to patient-centered interventional strategies. For example, the strategic use of health care navigators for adults diagnosed with pCNVs, many of whom have low literacy, may increase medication compliance and wellness behaviors. In addition, we propose that medicalizing NPDs through diagnosis and genetic counseling about a pCNV etiology may decrease the stigma of having a mental health disorder and lead to closer engagement of these patients with health care providers.12
This initial, broad look at selected medical aspects and health care resource utilization among adults with NPD-related pCNVs suggests that genetic information may add substantially to what can be captured from EHR clinical diagnostic data alone. These findings highlight the potential for genetic information–specifically, pCNVs−to contribute to our understanding of the determinants of health outcomes and inform prediction of high resource utilization in adults. The increased prevalence of NPDs in the pCNV cohort could be an important contributing factor to the observed higher rates of chronic medical conditions in this group. One potential limitation of this study relates to grouping different pCNVs in the analyses because some of the findings could be disproportionately influenced by specific pCNVs with known disease associations (eg, renal failure in 17q12 deletions). Additional research is needed to determine whether specific genetic diagnoses, symptom severity, visit types, or other utilization measures are driving the pCNV effect. If, as these findings suggest, individuals with NPD-associated pCNVs have poorer health and require disproportionate health care resources, early genetic diagnosis paired with patient-centered interventions may help to anticipate or prevent problems, improve outcomes, and reduce the associated economic burden.
Data availability:
Data in the manuscript were derived from aggregate, de-identified patient records, and analyzed under a full waiver of consent and Health Insurance Portability and Accountability Act authorization. As such, unique identifiers could not be generated for individual patient records for public sharing. Researchers who present proof of Institutional Review Board approval may request the aggregate data from the corresponding author.
Acknowledgments:
This work was supported by the National Institute of Mental Health of the National Institutes of Health, grant numbers U01MH119705 and R01MH074090.
Footnotes
Conflict of Interest: David H. Ledbetter is a scientific consultant for the following commercial entities: Dascena, Inc; Seven Bridges Genomics; Natera, Inc; MyOme, Inc; X-Therma, Inc. All other authors declare no conflicts of interest.
Ethics declaration: MyCode and the research outlined are approved by the Geisinger Institutional Review Board. Consent is obtained from participants or participants’ parent/legal guardian. Individual data included in this manuscript have been de-identified.
References
- 1.Tammimies K, Marshall CR, Walker S, et al. Molecular diagnostic yield of chromosomal microarray analysis and whole-exome sequencing in children with autism spectrum disorder. JAMA. 2015;314:895–903. [DOI] [PubMed] [Google Scholar]
- 2.Kirov G CNVs in neuropsychiatric disorders. Hum Mol Genet. 2015;24:45–49. [DOI] [PubMed] [Google Scholar]
- 3.Martin CL, Wain KE, Oetjens MT, et al. Identification of neuropsychiatric copy number variants in a health care system population. JAMA Psychiatry. 2020;77:1276–1285. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Crawford K, Bracher-Smith M, Owen D, et al. Medical consequences of pathogenic CNVs in adults: analysis of the UK Biobank. J Med Genet. 2019;56:131–138. [DOI] [PubMed] [Google Scholar]
- 5.Owen D, Bracher-Smith M, et al. Effects of pathogenic CNVs on physical traits in participants of the UK Biobank. BMC Genomics. 2018;19:867. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Mannik K, Magi R, Mace A, et al. Copy number variations and cognitive phenotypes in unselected populations. JAMA. 2015;313:2044–2054. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Stefansson H, Meyer-Lindenberg A, Steinberg S, et al. CNVs conferring risk of autism or schizophrenia affect cognition in controls. Nature. 2014;505:361–366. [DOI] [PubMed] [Google Scholar]
- 8.Singh T, Walters JTR, Johnstone M, et al. The contribution of rare variants to risk of schizophrenia in individuals with and without intellectual disability. Nat Genet. 2017;49:1167–1173. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Lowther C, Merico D, Costain G, et al. Impact of IQ on the diagnostic yield of chromosomal microarray in a community sample of adults with schizophrenia. Genome Med. 2017;9:105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Miller DT, Adam MP, Aradhya S, et al. Consensus statement: chromosomal microarray is a first-tier clinical diagnostic test for individuals with developmental disabilities or congenital anomalies. Am J Hum Genet. 2010;86:749–764. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Costain G, Lionel AC, Merico D, et al. Pathogenic rare copy number variants in community-based schizophrenia suggest a potential role for clinical microarrays. Hum Mol Genet. 2013;22:4485–4501. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Finucane BM, Ledbetter DH, Vorstman JA. Diagnostic genetic testing for neurodevelopmental psychiatric disorders: closing the gap between recommendation and clinical implementation. Curr Opin Genet Dev. 2021;68:1–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Srivastava S, Love-Nichols J A, Dies KA, et al. Meta-analysis and multidisciplinary consensus statement: exome sequencing is a first-tier clinical diagnostic test for individuals with neurodevelopmental disorders. Genet Med 2019;21:2413–2421. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Manickam K, McClain MR, Demmer LA, et al. Exome and genome sequencing for pediatric patients with congenital anomalies or intellectual disability: an evidence-based clinical guideline of the American College of Medical Genetics and Genomics (ACMG). Genet Med. 2021;doi: 10.1038/s41436-021-01242-6. Epub ahead of print. [DOI] [PubMed] [Google Scholar]
- 15.Valencia CA, Husami A, Holle J, et al. Clinical impact and cost-effectiveness of whole exome sequencing as a diagnostic tool: a pediatric center’s experience. Front Pediatr. 2015;3:67. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Fung WL, Butcher NJ, Costain G, et al. Practical guidelines for managing adults with 22q11.2 deletion syndrome. Genet Med. 2015;17:599–609. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Roehlen N, Hilger H, Stock F, et al. 17q12 deletion syndrome as a rare cause for diabetes mellitus type MODY5. J Clin Endocrinol Metab. 2018;103:3601–3610. [DOI] [PubMed] [Google Scholar]
- 18.Chung WK, Roberts TP, Sherr EH, Snyder LG, Spiro JE. 16p11.2 deletion syndrome. Curr Opin Genet Dev. 2021;68:49–56. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Finucane B, Savatt JM, Shimelis H, Girirajan S, Myers SM. Birt-Hogg-Dubé symptoms in Smith-Magenis syndrome include pediatric-onset pneumothorax. Am J Med Genet A. 2021;185:1922–1924. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.de Oliveira C, Cheng J, Vigod S, Rehm J, Kurdyak P. Patients with high mental health costs incur over 30 percent more costs than other high-cost patients. Health Aff (Millwood). 2016;35:36–43. [DOI] [PubMed] [Google Scholar]
- 21.Scott KM, Lim C, Al-Hamzawi A, et al. Association of mental disorders with subsequent chronic physical conditions: world mental health surveys from 17 countries. JAMA Psychiatry. 2016;73:150–158. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Niedzwiecki MJ, Sharma PJ, Kanzaria HK, McConville S, Hsia RY. Factors associated with emergency department use by patients with and without mental health diagnoses. JAMA Netw Open. 2018;1:e183528. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Hunt KA, Weber EJ, Showstack JA, Colby DC, Callaham ML. Characteristics of frequent users of emergency departments. Ann Emerg Med. 2006;48:1–8. [DOI] [PubMed] [Google Scholar]
- 24.Carey DJ, Fetterolf SN, Davis FD, et al. The Geisinger MyCode community health initiative: an electronic health record-linked biobank for precision medicine research. Genet Med. 2016;18:906–913. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40:373–383. [DOI] [PubMed] [Google Scholar]
- 26.Dewey FE, Murray MF, Overton JD, et al. Distribution and clinical impact of functional variants in 50,726 whole-exome sequences from the DiscovEHR study. Science. 2016;354:6319. [DOI] [PubMed] [Google Scholar]
- 27.Packer JS, Maxwell EK, O’Dushlaine C, et al. CLAMMS: a scalable algorithm for calling common and rare copy number variants from exome sequencing data. Bioinformatics. 2016;32:133–135. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.World Health Organization. ICD-10: international statistical classification of diseases and related health problems: tenth revision, 2nd ed. Geneva, Switzerland: World Health Organization; 2004. [Google Scholar]
- 29.Rosner B, Glynn RJ, Lee ML. Incorporation of clustering effects for the Wilcoxon rank sum test: a large-sample approach. Biometrics. 2003;59:1089–1098. [DOI] [PubMed] [Google Scholar]
- 30.Macé A, Tuke MA, Deelen P, et al. CNV-association meta-analysis in 191,161 European adults reveals new loci associated with anthropometric traits. Nat Commun. 2017;8:744. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.SAS Institute [SAS/ACCESS®]. Release 9.4. Cary, NC: SAS Institute, 2013. [Google Scholar]
- 32.R Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing, 2020. [Google Scholar]
- 33.Khan Y, Glazier RH, Moineddin R, Schull MJ. A population-based study of the association between socioeconomic status and emergency department utilization in Ontario, Canada. Acad Emerg Med. 2011;18:836–843. [DOI] [PubMed] [Google Scholar]
- 34.Wain KE, Tolwinski K, Palen E, et al. Population genomic screening for genetic etiologies of neurodevelopmental/psychiatric disorders demonstrates personal utility and positive participant responses. J Pers Med. In press. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Kelly E, Ivers N, Zawi R, et al. Patient navigators for people with chronic disease: protocol for a systematic review and meta-analysis. Syst Rev 2015;4:28. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Kalia SS, Adelman K, Bale SJ, et al. : Recommendations for reporting of secondary findings in clinical exome and genome sequencing, 2016 update (ACMG SF v2.0): a policy statement of the American College of Medical Genetics and Genomics. Genet Med. 2017;19:249–255. [DOI] [PubMed] [Google Scholar]
- 37.Cullen B, Smith DJ, Deary IJ, et al. The “cognitive footprint” of psychiatric and neurological conditions: cross-sectional study in the UK Biobank cohort. Acta Psychiatr Scand. 2017;135:593–605. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Edwards CH, Aas E, Kinge JM. Body mass index and lifetime healthcare utilization. BMC Health Serv Res. 2019;19:696. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Baumgart M, Snyder HM, Carrillo MC, Fazio S, Kim H, Johns H. Summary of the evidence on modifiable risk factors for cognitive decline and dementia: A population-based perspective. Alzheimers Dement. 2015;11:718–726. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
Data in the manuscript were derived from aggregate, de-identified patient records, and analyzed under a full waiver of consent and Health Insurance Portability and Accountability Act authorization. As such, unique identifiers could not be generated for individual patient records for public sharing. Researchers who present proof of Institutional Review Board approval may request the aggregate data from the corresponding author.