Using Administrative Data to Ascertain True Cases of Muscular Dystrophy: Rare Disease Surveillance

Development and electronic health record validation of an algorithm for identifying patients with Duchenne muscular dystrophy in US administrative claims

BACKGROUND: Muscular dystrophies (MDs) comprise a heterogenous group of genetically inherited conditions characterized by progressive muscle weakness and increasing disability. The lack of separate diagnosis codes for Duchenne MD (DMD) and Becker MD, 2 of the most common forms of MD, has limited the conduct of DMD-specific real-world studies. OBJECTIVE: To develop and validate administrative claims-based algorithms for identifying patients with DMD and capturing their nonambulatory and ventilation-dependent status. METHODS: This was a retrospective cohort study using the statistically deidentified Optum Market Clarity Database (including patient claims linked with electronic health records [EHRs] data) to develop and validate the following algorithms: DMD diagnosis, nonambulatory status, and ventilation-dependent status. The initial study sample consisted of US patients in the database who had a diagnosis code for Duchenne/Becker MD (DBMD) between October 1, 2018, and September 30, 2020, who were male, aged 40 years or younger on their first DBMD diagnosis, and met continuous enrollment and 1-day minimal clinical activities requirement in a 12-month measurement period between October 1, 2017, and September 30, 2020. The algorithms, developed by a cross-functional team of DMD specialists (including patient advocates), were based on administrative claims data with International Classification of Diseases, Tenth Revision, Clinical Modifications coding, using information of diagnosis codes for DBMD, sex, age, treatment, and disease severity (eg, evidence of ambulation assistance/support and/or evidence of ventilation support or dependence). Patients who met each algorithm and had EHR notes available were then validated against structured fields and unstructured provider notes from their own linked EHR to confirm patients' DMD diagnoses, nonambulatory status, and ventilation-dependent status. Algorithm performance was assessed by positive predictive value with 95% CIs. RESULTS: A total of 1,300 patients were included in the initial study sample. Of these, EHR were available and reviewed for 303 patients. The mean age of the 303 patients was 14.8 years, with 61.7% being non-Hispanic White. A majority had a Charlson comorbidity index score of 0 (59.4%) or 1-2 (27.7%). Positive predictive value (95% CI) was 91.6% (85.8%-95.6%) for the DMD diagnosis algorithm, 88.4% (80.2%-94.1%) for the nonambulatory status algorithm, and 77.8% (62.9%-88.8%) for the ventilation-dependent status algorithm. CONCLUSIONS: This work provides the means to more accurately identify patients with DMD from administrative claims data without a specific diagnosis code. The algorithms validated in this study can be applied to assess treatment effectiveness and other outcomes among patients with DMD treated in clinical practice. DISCLOSURES: This study was funded by Pfizer, which contracted with Optum to perform the study and provide medical writing assistance. Ms Schrader reports being an employee of Parent Project Muscular Dystrophy. Mr Posner reports being an employee and stockholder of Pfizer and receiving support from Pfizer for attending conferences not related to this manuscript. Dr Dorling reports being an employee and stockholder of Pfizer at the time the study was conducted and is a current employee of Chiesi USA, Inc. Ms Senerchia reports being an employee of Optum and owning stock in Pfizer and UnitedHealth Group, the parent company of Optum. Dr Chen reports being an employee and stockholder of Pfizer. Ms Beaverson reports being an employee of Pfizer and owning stock in Pfizer and Amicus Therapeutics. Dr Seare reports being an employee of Optum at the time the study was conducted. Dr Garnier and Ms Merla report being employees of Pfizer. Ms Walker reports being an employee of Optum. Dr Alvir reports being an employee and stockholder of Pfizer. Dr Mahn reports being an employee and stockholder of Pfizer. Dr Zhang reports being an employee of Optum. Ms Landis reports being an employee of Optum. Ms Buikema reports being an employee of Optum and holding stock in UnitedHealth Group, the parent company of Optum.

Leveraging machine learning to identify acute myeloid leukemia patients and their chemotherapy regimens in an administrative database

BACKGROUND

Administrative datasets are useful for identifying rare disease cohorts such as pediatric acute myeloid leukemia (AML). Previously, cohorts were assembled using labor-intensive, manual reviews of patients' longitudinal chemotherapy data.

METHODS

We utilized a two-step machine learning (ML) method to (i) identify pediatric patients with newly diagnosed AML, and (ii) among the identified AML patients, their chemotherapy courses, in an administrative/billing database. Using 2558 patients previously manually reviewed, multiple ML algorithms were derived from 75% of the study sample, and the selected model was tested in the remaining hold-out sample. The selected model was also applied to assemble a new pediatric AML cohort and further assessed in an external validation, using a standalone cohort established by manual chart abstraction.

RESULTS

For patient identification, the selected Support Vector Machine model yielded a sensitivity of 0.97 and a positive predictive value (PPV) of 0.97 in the hold-out test sample. For course-specific chemotherapy regimen and start date identification, the selected Random Forest model yielded overall PPV greater than or equal to 0.88 and sensitivity greater than or equal to 0.86 across all courses in the test sample. When applied to new cohort assembly, ML identified 3016 AML patients with 10,588 treatment courses. In the external validation subset, PPV was greater than or equal to 0.75 and sensitivity was greater than or equal to 0.82 for patient identification, and PPV was greater than or equal to 0.93 and sensitivity was greater than or equal to 0.94 for regimen identifications.

CONCLUSION

A carefully designed ML model can accurately identify pediatric AML patients and their chemotherapy courses from administrative databases. This approach may be generalizable to other diseases and databases.

Magnitude of the Potential Screening Gap for Fabry Disease in Manitoba: A Population-Based Retrospective Cohort Study

Background:

Fabry disease is a rare disorder caused by the deficient activity of α-galactosidase A (GLA) that often leads to organ damage. Fabry disease can be treated with enzyme replacement or pharmacological therapy, but due to its rarity and nonspecific manifestations, it often goes undiagnosed. Mass screening for Fabry disease is impractical; however, a targeted screening program for high-risk individuals may uncover previously unknown cases.

Objective:

Our objective was to use population-level administrative health databases to identify patients at high risk of Fabry disease.

Design:

Retrospective cohort study.

Setting:

Population-level health administrative databases housed at the Manitoba Centre for Health Policy.

Patients:

All residents of Manitoba, Canada, between 1998 and 2018.

Measurements:

We ascertained the evidence of GLA testing in a cohort of patients at high risk of Fabry disease.

Methods:

Individuals without a hospitalization or prescription indicative of Fabry disease were included if they had evidence of 1 of 4 high-risk conditions for Fabry disease: (1) ischemic stroke <45 years of age, (2) idiopathic hypertrophic cardiomyopathy, (3) proteinuric chronic kidney disease or kidney failure of unknown cause, or (4) peripheral neuropathy. Patients were excluded if they had known contributing factors to these high-risk conditions. Those who remained and had no prior GLA testing were assigned a 0% to 4.2% probability of having Fabry disease depending on their high-risk condition and sex.

Results:

After applying exclusion criteria, 1386 individuals were identified as having at least 1 high-risk clinical condition for Fabry disease in Manitoba. There were 416 GLA tests conducted during the study period, and of those, 22 were conducted in individuals with at least 1 high-risk condition. This leaves a screening gap of 1364 individuals with a high-risk clinical condition for Fabry disease in Manitoba who have not been tested. At the end of the study period, 932 of those individuals were still alive and residing in Manitoba, and if screened today, we expect between 3 and 18 would test positive for Fabry disease.

Limitations:

The algorithms we used to identify our patients have not been validated elsewhere. Diagnoses of Fabry disease, idiopathic hypertrophic cardiomyopathy, and peripheral neuropathy were only available via hospitalizations and not physician claims. We were only able to capture GLA testing processed through public laboratories. Patients identified to be at high risk of Fabry disease by the algorithm did not undergo GLA testing due to a clinical rationale that we were unable to capture.

Conclusions:

Administrative health databases may be a useful tool to identify patients at higher risk of Fabry disease or other rare conditions. Further directions include designing a program to screen high-risk individuals for Fabry disease as identified by our administrative data algorithms.

Heterogeneity in Autism Spectrum Disorder Case-Finding Algorithms in United States Health Administrative Database Analyses

Strengthening systems of care to meet the needs of individuals with autism spectrum disorder (ASD) is of growing importance. Administrative data provide advantages for research and planning purposes, including large sample sizes and the ability to identify enrollment in insurance coverage and service utilization of individuals with ASD. Researchers have employed varying strategies to identify individuals with ASD in administrative data. Differences in these strategies can limit the comparability of results across studies. This review describes implications of the varying strategies that have been employed to identify individuals with ASD in US claims databases, with consideration of the strengths and limitations of each approach.

Geographic Barriers Drive Disparities in Specialty Center Access for Older Adults with Huntington's Disease

BACKGROUND

Huntington's Disease Society of America Centers of Excellence (HDSA COEs) are primary hubs for Huntington's disease (HD) research opportunities and accessing new treatments. Data on the extent to which HDSA COEs are accessible to individuals with HD, particularly those older or disabled, are lacking.

OBJECTIVE

To describe persons with HD in the U.S. Medicare program and characterize this population by proximity to an HDSA COE.

METHODS

We conducted a cross-sectional study of Medicare beneficiaries ages ≥65 with HD in 2017. We analyzed data on benefit entitlement, demographics, and comorbidities. QGis software and Google Maps Interface were employed to estimate the distance from each patient to the nearest HDSA COE, and the proportion of individuals residing within 100 miles of these COEs at the state level.

RESULTS

Among 9,056 Medicare beneficiaries with HD, 54.5% were female, 83.0% were white; 48.5% were ≥65 years, but 64.9% originally qualified for Medicare due to disability. Common comorbidities were dementia (32.4%) and depression (35.9%), and these were more common in HD vs. non-HD patients. Overall, 5,144 (57.1%) lived within 100 miles of a COE. Race/ethnicity, sex, age, and poverty markers were not associated with below-average proximity to HDSA COEs. The proportion of patients living within 100 miles of a center varied from < 10% (16 states) to > 90% (7 states). Most underserved states were in the Mountain and West Central divisions.

CONCLUSION

Older Medicare beneficiaries with HD are frequently disabled and have a distinct comorbidity profile. Geographical, rather than sociodemographic factors, define the HD population with limited access to HDSA COEs.

Direct costs of adhering to selected Duchenne muscular dystrophy care considerations: estimates from a Midwestern state

INTRODUCTION/AIMS

The multidisciplinary Duchenne muscular dystrophy (DMD) Care Considerations were developed to standardize care and improve outcomes. We provide cumulative cost estimates for selected key preventive (i.e., excluding new molecular therapies and acute care) elements of the care considerations in eight domains (neuromuscular, rehabilitation, respiratory, cardiac, orthopedic, gastrointestinal, endocrine, psychosocial management) independent of completeness of uptake or provision of non-preventive care.

METHODS

We used de-identified insurance claims data from a large Midwestern commercial health insurer during 2018. We used Current Procedural Terminology and National Drug codes to extract unit costs for clinical encounters representing key preventive elements of the DMD Care Considerations. We projected per-patient cumulative costs from ages 5 to 25 years for these elements by multiplying a schedule of recommended frequencies of preventive services by unit costs in 2018 US dollars.

RESULTS

Assuming a diagnosis at age 5 years, independent ambulation until age 11, and survival until age 25, we estimated 670 billable clinical events. The 20-year per-patient cumulative cost was $174,701 with prednisone ($2.3 million with deflazacort) and an expected total of $12,643 ($29,194) for out-of-pocket expenses associated with those events and medications.

DISCUSSION

Standardized monitoring of disease progression and treatments may reduce overall costs of illness. Costs associated with these services would be needed to quantify potential savings. Our approach demonstrates a method to estimate costs associated with implementation of preventive care schedules.

Characterizing demographics, comorbidities, and costs of care among populations with Duchenne muscular dystrophy with Medicaid and commercial coverage

BACKGROUND: Duchenne muscular dystrophy (DMD) is a severe X-linked progressive neurodegenerative disease characterized by loss of ambulation, cardiomyopathy, respiratory insufficiency, and early mortality. Few data are available that describe the direct medical costs among patients with DMD in the United States. OBJECTIVE: To characterize the demographics, comorbidity burden, and direct monthly costs of care among patients with DMD with Medicaid and with commercial insurance coverage. METHODS: IBM MarketScan Commercial and Multi-State Medicaid claims (2013-2018) were used to identify males aged 30 years or under with diagnostic codes for muscular dystrophy or DMD; additional exclusion criteria were applied to identify those with probable DMD. Baseline characteristics and comorbidities were tabulated. The frequency of health care resource use and median (interquartile range [IQR]) monthly costs (in 2018 USD) were estimated from those with at least 12 months of continuous follow-up. RESULTS: Median (IQR) baseline ages were similar between the Medicaid (14 [9-20] years; n = 2,007) and commercial (15 [9-21] years; n = 1,964) DMD cohorts. The frequency of comorbidities over the period was slightly higher with those on Medicaid. The median duration of follow-up was 3.1 years among members of the Medicaid DMD cohort and 1.7 years among the commercial DMD cohort. Median monthly resource use was generally higher among the Medicaid DMD cohort; nonetheless, median (IQR) monthly costs were similar at $1,735 ($367-$5,281) for the Medicaid DMD cohort vs $1,883 ($657-$6,796) for the commercial DMD cohort. CONCLUSIONS: The demographic characteristics and median direct medical costs were similar between patients with commercial vs Medicaid coverage, even though patients with Medicaid coverage had higher resource use. Despite challenges in definitively identifying DMD patients using claims data, these findings help characterize contemporary DMD populations in the United States and the related direct economic burden to the payer. DISCLOSURES: This study was funded by Sarepta Therapeutics, Inc. Klimchak and Gooch are employees of Sarepta Therapeutics Inc. Szabo, Qian, and Popoff are employees of Broadstreet HEOR, which received funds from Sarepta Therapeutics, Inc., for work on this study. Iannaccone has received research funding or consulting fees from Avexis, Biogen, Fibrogen, Mallinkrodt, Regeneron, Sarepta Therapeutics, Inc., Scholar Rock, PTC Therapeutics, Pfizer, MDA, CureSMA, NIH, Genentech-Roche, and BCBS. Publication of the study results was not contingent on the sponsor's approval or censorship of the manuscript. Information from this study was presented, in part, at the AMCP Virtual Annual Meeting, April 21-24, 2020.

Assessment of the Accuracy of Identification of Selected Disabilities and Conditions in Hospital Discharge Data for Pregnant Women

BACKGROUND

Linked birth certificate-hospital discharge records are a valuable resource for examining pregnancy outcomes among women with disability conditions. Few studies relying on these data have been able to assess the accuracy of identification of preexisting disability conditions. We assessed the accuracy of International Classification of Diseases version 9 (ICD9) codes for identifying selected physical, sensory, and intellectual conditions that may result in disability. As ICD9 codes were utilized until recently in most states, this information is useful to inform analyses with these records.

METHODS

We reviewed 280 of 311 (90%) medical records of pregnant women with disabilities based on ICD9 codes and 390 of 8,337 (5%) records of pregnant women without disabilities who had deliveries at a large university medical center. We estimated sensitivity, specificity, and positive predictive values (PPV) using the medical record as gold standard. We adjusted for verification bias using inverse probability weighting and imputation.

RESULTS

The estimated sensitivity of ICD9 codes to identify women with disabilities with deliveries 2009-2012 was 44%; PPV was 98%, improving over time. Although sensitivity was <50% for some conditions, PPVs were 87%-100% for all conditions except intellectual disability (67%). Many physical conditions had complete verification and no underreporting.

CONCLUSIONS

These results are helpful for new studies using historical data comparing outcomes among women with and without these conditions and to inform interpretation of results from earlier studies. Assessment of the accuracy of disabilities as identified by ICD version 10 codes is warranted.

Muscular Dystrophy Surveillance, Tracking, and Research Network pilot: Population-based surveillance of major muscular dystrophies at four U.S. sites, 2007-2011

BACKGROUND

For 10 years, the Muscular Dystrophy Surveillance, Tracking, and Research Network (MD STARnet) conducted surveillance for Duchenne and Becker muscular dystrophy (DBMD). We piloted expanding surveillance to other MDs that vary in severity, onset, and sources of care.

METHODS

Our retrospective surveillance included individuals diagnosed with one of nine eligible MDs before or during the study period (January 2007-December 2011), one or more health encounters, and residence in one of four U.S. sites (Arizona, Colorado, Iowa, or western New York) at any time within the study period. We developed case definitions, surveillance protocols, and software applications for medical record abstraction, clinical review, and data pooling. Potential cases were identified by International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes 359.0, 359.1, and 359.21 and International Classification of Diseases, Tenth Revision (ICD-10) codes G71.0 and G71.1. Descriptive statistics were compared by MD type. Percentage of MD cases identified by each ICD-9-CM code was calculated.

RESULTS

Of 2,862 cases, 32.9% were myotonic, dystrophy 25.8% DBMD, 9.7% facioscapulohumeral MD, and 9.1% limb-girdle MD. Most cases were male (63.6%), non-Hispanic (59.8%), and White (80.2%). About, half of cases were genetically diagnosed in self (39.1%) or family (6.2%). About, half had a family history of MD (48.9%). The hereditary progressive MD code (359.1) was the most common code for identifying eligible cases. The myotonic code (359.21) identified 83.4% of eligible myotonic dystrophy cases (786/943).

CONCLUSIONS

MD STARnet is the only multisite, population-based active surveillance system available for MD in the United States. Continuing our expanded surveillance will contribute important epidemiologic and health outcome information about several MDs.

Identifying Algorithms to Improve the Accuracy of Unverified Diagnosis Codes for Birth Defects

OBJECTIVES

We identified algorithms to improve the accuracy of passive surveillance programs for birth defects that rely on administrative diagnosis codes for case ascertainment and in situations where case confirmation via medical record review is not possible or is resource prohibitive.

METHODS

We linked data from the 2009-2011 Florida Birth Defects Registry, a statewide, multisource, passive surveillance program, to an enhanced surveillance database with selected cases confirmed through medical record review. For each of 13 birth defects, we calculated the positive predictive value (PPV) to compare the accuracy of 4 algorithms that varied case definitions based on the number of diagnoses, medical encounters, and data sources in which the birth defect was identified. We also assessed the degree to which accuracy-improving algorithms would affect the Florida Birth Defects Registry's completeness of ascertainment.

RESULTS

The PPV generated by using the original Florida Birth Defects Registry case definition (ie, suspected cases confirmed by medical record review) was 94.2%. More restrictive case definition algorithms increased the PPV to between 97.5% (identified by 1 or more codes/encounters in 1 data source) and 99.2% (identified in >1 data source). Although PPVs varied by birth defect, alternative algorithms increased accuracy for all birth defects; however, alternative algorithms also resulted in failing to ascertain 58.3% to 81.9% of cases.

CONCLUSIONS

We found that surveillance programs that rely on unverified diagnosis codes can use algorithms to dramatically increase the accuracy of case finding, without having to review medical records. This can be important for etiologic studies. However, the use of increasingly restrictive case definition algorithms led to a decrease in completeness and the disproportionate exclusion of less severe cases, which could limit the widespread use of these approaches.

Utilizing Combined Claims and Clinical Datasets for Research Among Potential Cases of Rare Diseases

With data quality issues with administrative claims and medically derived datasets, a dataset derived from a combination of sources may be more effective for research. The purposes of this article is to link an EMR-based data warehouse with state administrative data to study individuals with rare diseases; to describe and compare their characteristics; and to explore research with the data. These methods included subjects with diagnosis codes for one of three rare diseases from the years 2009-2014; Spina Bifida, Muscular Dystrophy, and Fragile X Syndrome. The results from the combined data provides additional information that each dataset, by itself, would not contain. The simultaneous examination of data such as race/ethnicity, physician and other outpatient visit data, charges and payments, and overall utilization was possible in the combined dataset. It is also discussed that combining such datasets can be a useful tool for the study of populations with rare diseases.

Capture-recapture methodology to study rare conditions using surveillance data for fragile X syndrome and muscular dystrophy

BACKGROUND

Rare conditions can be catastrophic for families and the implications for public health can be substantial. Our study compared basic surveillance through active medical record review with a linked administrative data file to assess the number of cases of two rare conditions, fragile X syndrome (FXS) and muscular dystrophy (MD) in a population.

METHODS

Two methods of data collection were used to collect information from five counties comprising two standard metropolitan statistical areas of South Carolina. The passive system relied mostly on health claims data using ICD-9 CM diagnostic codes. The active system relied on a nurse abstracting records from a list of all licensed physicians with specialties in neurology, orthopedics, and genetics.

RESULTS

There were 141 FXS cases and 348 MD cases that met the case definitions using active surveillance. Additional cases were found for both conditions but they were determined to not be true cases. After linking the actively collected MD and FXS cases to passive datasets, we found that the estimated total numbers of cases were similar to using capture-recapture analysis; the positive predictive values for cases identified in the passive system were 56.6% for MD and 75.7% for FXS.

CONCLUSIONS

Applying capture-recapture methods to passively collected surveillance data for rare health conditions produced an estimate of the number of true cases that was similar to that obtained through active data collection.