Scalable and interpretable alternative to chart review for phenotype evaluation using standardized structured data from electronic health records

OBJECTIVES

Chart review as the current gold standard for phenotype evaluation cannot support observational research on electronic health records and claims data sources at scale. We aimed to evaluate the ability of structured data to support efficient and interpretable phenotype evaluation as an alternative to chart review.

MATERIALS AND METHODS

We developed Knowledge-Enhanced Electronic Profile Review (KEEPER) as a phenotype evaluation tool that extracts patient's structured data elements relevant to a phenotype and presents them in a standardized fashion following clinical reasoning principles. We evaluated its performance (interrater agreement, intermethod agreement, accuracy, and review time) compared to manual chart review for 4 conditions using randomized 2-period, 2-sequence crossover design.

RESULTS

Case ascertainment with KEEPER was twice as fast compared to manual chart review. 88.1% of the patients were classified concordantly using charts and KEEPER, but agreement varied depending on the condition. Missing data and differences in interpretation accounted for most of the discrepancies. Pairs of clinicians agreed in case ascertainment in 91.2% of the cases when using KEEPER compared to 76.3% when using charts. Patient classification aligned with the gold standard in 88.1% and 86.9% of the cases respectively.

CONCLUSION

Structured data can be used for efficient and interpretable phenotype evaluation if they are limited to relevant subset and organized according to the clinical reasoning principles. A system that implements these principles can achieve noninferior performance compared to chart review at a fraction of time.

Traumatic Brain Injury Classification Variability During the Afghanistan/Iraq Conflicts: Surveillance, Clinical, Research, and Policy Implications

OBJECTIVE

Challenges associated with case ascertainment of traumatic brain injuries (TBIs) sustained during the Afghanistan/Iraq military operations have been widespread. This study was designed to examine how the prevalence and severity of TBI among military members who served during the conflicts were impacted when a more precise classification of TBI diagnosis codes was compared with the Department of Defense Standard Surveillance Case-Definition (DoD-Case-Definition).

SETTING

Identification of TBI diagnoses in the Department of Defense's Military Health System from October 7, 2001, until December 31, 2019.

PARTICIPANTS

Military members with a TBI diagnosis on an encounter record during the study window.

DESIGN

Descriptive observational study to evaluate the prevalence and severity of TBI with regard to each code set (ie, the DoD-Case-Definition and the more precise set of TBI diagnosis codes). The frequencies of index TBI severity were compared over time and further evaluated against policy changes.

MAIN MEASURES

The more precise TBI diagnosis code set excludes the following: (1) DoD-only extender codes, which are not used in other healthcare settings; and (2) nonprecise TBI codes, which include injuries that do not necessarily meet TBI diagnostic criteria.

RESULTS

When comparing the 2 TBI classifications, the DoD-Case-Definition captured a higher prevalence of TBIs; 38.5% were classified by the DoD-Case-Definition only (>164 000 military members). 73% of those identified by the DoD-Case-Definition only were diagnosed with nonprecise TBI codes only, with questionable specificity as to whether a TBI occurred.

CONCLUSION

We encourage the field to reflect on decisions made pertaining to TBI case ascertainment during the height of the conflicts. Efforts focused on achieving consensus regarding TBI case ascertainment are recommended. Doing so will allow the field to be better prepared for future conflicts, and improve surveillance, screening, and diagnosis in noncombat settings, as well as our ability to understand the long-term effects of TBI.

Identifying cases of chronic pain using health administrative data: A validation study

Background

Most prevalence estimates of chronic pain are derived from surveys and vary widely, both globally (2%–54%) and in Canada (6.5%–44%). Health administrative data are increasingly used for chronic disease surveillance, but their validity as a source to ascertain chronic pain cases is understudied.

Aim

The aim of this study was to derive and validate an algorithm to identify cases of chronic pain as a single chronic disease using provincial health administrative data.

Methods

A reference standard was developed and applied to the electronic medical records data of a Newfoundland and Labrador general population sample participating in the Canadian Primary Care Sentinel Surveillance Network. Chronic pain algorithms were created from the administrative data of patient populations with chronic pain, and their classification performance was compared to that of the reference standard via statistical tests of selection accuracy.

Results

The most performant algorithm for chronic pain case ascertainment from the Medical Care Plan Fee-for-Service Physicians Claims File was one anesthesiology encounter ever recording a chronic pain clinic procedure code OR five physician encounter dates recording any pain-related diagnostic code in 5 years with more than 183 days separating at least two encounters. The algorithm demonstrated 0.703 (95% confidence interval [CI], 0.685–0.722) sensitivity, 0.668 (95% CI, 0.657–0.678) specificity, and 0.408 (95% CI, 0.393–0.423) positive predictive value. The chronic pain algorithm selected 37.6% of a Newfoundland and Labrador provincial cohort.

Conclusions

A health administrative data algorithm was derived and validated to identify chronic pain cases and estimate disease burden in residents attending fee-for-service physician encounters in Newfoundland and Labrador.

The current scope of healthcare-associated infection surveillance activities in hospitalized immunocompromised patients: a systematic review

BACKGROUND

Immunocompromised patients are at increased risk of acquiring healthcare-associated infections (HAIs) and often require specialized models of care. Surveillance of HAIs is essential for effective infection-prevention programmes. However, little is known regarding standardized or specific surveillance methods currently employed for high-risk hospitalized patients.

METHODS

A systematic review adopting a narrative synthesis approach of published material between 1 January 2000 and 31 March 2018 was conducted. Publications describing the application of traditional and/or electronic surveillance of HAIs in immunocompromised patient settings were identified from the Ovid MEDLINE®, Ovid Embase® and Elsevier Scopus® search engines [PROSPERO international prospective register of systematic reviews (registration ID: CRD42018093651)].

RESULTS

In total, 2708 studies were screened, of whom 17 fulfilled inclusion criteria. Inpatients diagnosed with haematological malignancies were the most-represented immunosuppressed population. The majority of studies described manual HAI surveillance utilizing internationally accepted definitions for infection. Chart review of diagnostic and pathology reports was most commonly employed for case ascertainment. Data linkage of disparate datasets was performed in two studies. The most frequently monitored infections were bloodstream infections and invasive fungal disease. No surveillance programmes applied risk adjustment for reporting surveillance outcomes.

CONCLUSIONS

Targeted, tailored monitoring of HAIs in high-risk immunocompromised settings is infrequently reported in current hospital surveillance programmes. Standardized surveillance frameworks, including risk adjustment and timely data dissemination, are required to adequately support infection-prevention programmes in these populations.

Development and Evaluation of a Prediction Model for Ascertaining Rheumatic Heart Disease Status in Administrative Data

Background

Previous research has raised substantial concerns regarding the validity of the International Statistical Classification of Diseases and Related Health Problems (ICD) codes (ICD-10 I05-I09) for rheumatic heart disease (RHD) due to likely misclassification of non-rheumatic valvular disease (non-rheumatic VHD) as RHD. There is currently no validated, quantitative approach for reliable case ascertainment of RHD in administrative hospital data.

Methods

A comprehensive dataset of validated Australian RHD cases was compiled and linked to inpatient hospital records with an RHD ICD code (2000-2018, n=7555). A prediction model was developed based on a generalized linear mixed model structure considering an extensive range of demographic and clinical variables. It was validated internally using randomly selected cross-validation samples and externally. Conditional optimal probability cutpoints were calculated, maximising discrimination separately for high-risk versus low-risk populations.

Results

The proposed model reduced the false-positive rate (FPR) from acute rheumatic fever (ARF) cases misclassified as RHD from 0.59 to 0.27; similarly for non-rheumatic VHD from 0.77 to 0.22. Overall, the model achieved strong discriminant capacity (AUC: 0.93) and maintained a similar robust performance during external validation (AUC: 0.88). It can also be used when only basic demographic and diagnosis data are available.

Conclusion

This paper is the first to show that not only misclassification of non-rheumatic VHD but also of ARF as RHD yields substantial FPRs. Both sources of bias can be successfully addressed with the proposed model which provides an effective solution for reliable RHD case ascertainment from hospital data for epidemiological disease monitoring and policy evaluation.

Completeness of case ascertainment in Swedish hospital-based stroke registers

BACKGROUND

There is a worldwide development toward using data from hospital-based stroke registers to estimate epidemiological trends. However, incomplete case ascertainment may cause selection bias. We examined the completeness of case ascertainment and selection bias in two hospital-based Swedish stroke registers.

METHODS

First-ever stroke cases between March 2015 and February 2016 in the catchment area of Skåne University Hospital, Lund, Sweden, were included from multiple overlapping sources: two hospital-based stroke registers, Riksstroke-Lund and Lund Stroke Register (LSR); local outpatient and inpatient registers; primary care registers; and autopsy registers. The resulting population-based cohort was used as reference to assess completeness of case ascertainment and patient characteristics in Riksstroke-Lund and LSR.

RESULTS

In total, 400 stroke patients were identified. Riksstroke-Lund detected 328 (82%) patients, whereas LSR detected 363 (91%). Patients undetected by hospital-based registers had higher 28-day case fatality than those detected (44% vs 9%; P = .001). Patients only detected in primary care (n = 11) more often lived in healthcare facilities compared with those detected by hospital-based registers (57% vs 7%; P = .001). Patients not detected by Riksstroke-Lund, but detected by population-based sources, had less severe strokes (median NIHSS 3 vs 5; P = .013).

CONCLUSIONS

Some first-ever stroke patients, such as those with high early case fatality and those with mild stroke, may go undetected with hospital-based screening used in clinical stroke registers. This can result in selection bias due to not identifying specific groups of patients including some with high early case fatality and those living in healthcare facilities.

Case ascertainment in active paediatric surveillance systems: a report from the British Paediatric Surveillance Unit Ascertainment Group

The British Paediatric Surveillance Unit (BPSU) conducts surveillance of rare paediatric conditions using active, or prospective, case finding. The reliability of estimates of incidence, which is the primary outcome of public health importance, depends on ascertainment being as near complete as possible. This paper reviews evidence of the completeness of ascertainment in recent surveillance studies run through the BPSU. Ascertainment varied between 49% and 94% depending on the study. These are upper estimates. This was the basis of a discussion on barriers and facilitators of ascertainment which we have separated into factors related to the condition, factors related to the study methods, factors related to the study team and factors related to the surveillance system infrastructure. This leads to a series of recommendations to ensure continuing high levels of ascertainment in active surveillance studies.

Cancer Incidence in Older Adults in the United States: Characteristics, Specificity, and Completeness of the Data

INTRODUCTION

The number of cancer cases in the United States continues to grow as the number of older adults increases. Accurate, reliable and detailed incidence data are needed to respond effectively to the growing human costs of cancer in an aging population. The purpose of this study was to examine the characteristics of incident cases and evaluate the impact of death-certificate-only (DCO) cases on cancer incidence rates in older adults.

METHODS

Using data from 47 cancer registries and detailed population estimates from the Surveillance, Epidemiology and End Results (SEER) Program, we examined reporting sources, methods of diagnosis, tumor characteristics, and calculated age-specific incidence rates with and without DCO cases in adults aged 65 through ≥95 years, diagnosed 2011 through 2015, by sex and race/ethnicity.

RESULTS

The percentage of cases (all cancers combined) reported from a hospital decreased from 90.6% (ages 65-69 years) to 69.1% (ages ≥95 years) while the percentage of DCO cases increased from 1.1% to 19.6%. Excluding DCO cases, positive diagnostic confirmation decreased as age increased from 96.8% (ages 65-69 years) to 69.2% (ages ≥95 years). Compared to incidence rates that included DCO cases, rates in adults aged ≥95 years that excluded DCO cases were 41.5% lower in Black men with prostate cancer and 29.2% lower in Hispanic women with lung cancer.

DISCUSSION

Loss of reported tumor specificity with age is consistent with fewer hospital reports. However, the majority of cancers diagnosed in older patients, including those aged ≥95 years, were positively confirmed and were reported with known site, histology, and stage information. The high percentage of DCO cases among patients aged ≥85 years suggests the need to explore additional sources of follow-back to help possibly identify an earlier incidence report. Interstate data exchange following National Death Index linkages may help registries identify and remove erroneous DCO cases from their databases.

Under-reporting of sudden unexpected death in epilepsy

The identification and characterization of sudden unexpected deaths in epilepsy (SUDEP) may be improved, helping to optimize prevention and intervention. We set out to assess the frequency and demographic and clinical characteristics of SUDEP cases in a sudden death cohort. All out-of-hospital deaths were investigated from March 1, 2013 to February 28, 2015 in Wake County, NC, attended by the Emergency Medical Services. Cases were screened and adjudicated by three physicians to identify sudden death cases from any cause among free-living adults, aged 18-64. In total, 399 sudden death victims were identified during this two-year period. Seizure history, demographic and clinical characteristics, and healthcare utilization patterns were assessed from death records, emergency response scene reports, and medical records. Sudden death cases with a history of seizures were summarized by an experienced chart abstractor (SC) and adjudicated by an experienced neurologist (OD). We then compared demographic and clinical characteristics and healthcare utilization patterns of neurologist-identified SUDEP cases to other sudden death victims in our population-based registry of sudden death from any cause. SUDEP accounted for 5.3% of sudden deaths. However, seizures or complications of seizures were only considered the primary cause of death on death certificates in 1.5% of sudden deaths. SUDEP cases were more likely to have a history of alcohol abuse. Mental health disorders and a low level of medication compliance and healthcare utilization were common among SUDEP victims. SUDEP accounts for approximately 5.3% of sudden deaths from any cause in individuals aged between 18 and 64. Death certificates underestimate the burden of sudden death in epilepsy, attributing only 1.5% of sudden deaths to seizures or complications of seizures. Accurate documentation of epileptic disorders on death certificates is essential for the surveillance of SUDEP. Further, interventions that promote better use of medical services and patient engagement with healthy living practices may reduce sudden deaths in epilepsy.

Development and validation of novel algorithms to identify patients with inflammatory bowel diseases in Israel: an epi-IIRN group study

Background

Before embarking on administrative research, validated case ascertainment algorithms must be developed. We aimed at developing algorithms for identifying inflammatory bowel disease (IBD) patients, date of disease onset, and IBD type (Crohn's disease [CD] vs ulcerative colitis [UC]) in the databases of the four Israeli Health Maintenance Organizations (HMOs) covering 98% of the population.

Methods

Algorithms were developed on 5,131 IBD patients and 2,072 controls, following independent chart review (60% CD and 39% UC). We reviewed 942 different combinations of clinical parameters aided by mathematical modeling. The algorithms were validated on an independent cohort of 160,000 random subjects.

Results

The combination of the following variables achieved the highest diagnostic accuracy: IBD-related codes, alone if more than five to six codes or combined with purchases of IBD-related medications (at least three purchases or ≥3 months from the first to last purchase) (sensitivity 89%, specificity 99%, positive predictive value [PPV] 92%, negative predictive value [NPV] 99%). A look-back period of 2-5 years (depending on the HMO) without IBD-related codes or medications best determined the date of diagnosis (sensitivity 83%, specificity 68%, PPV 82%, NPV 70%). IBD type was determined by the majority of CD/UC codes of the three recent contacts or the most recent when less than three contacts were recorded (sensitivity 92%, specificity 97%, PPV 97%, NPV 92%). Applying these algorithms, a total of 38,291 IBD patients were residing in Israel, corresponding to a prevalence rate of 459/100,000 (0.46%).

Conclusion

The application of the validated algorithms to Israel's administrative databases will now create a large and accurate ongoing population-based cohort of IBD patients for future administrative studies.

The Impact of Hospital Discharge Linkage on Case Ascertainment of Brain Tumors in the Alberta Cancer Registry, 2010-2015

BACKGROUND

Concern has been raised regarding the underreporting of nonmalignant central nervous system tumors. This study addressed this issue with 2 objectives: (1) evaluate the impact of linkage with hospital discharges, as recorded in the Discharge Abstract Database (DAD), on supplementing case ascertainment for brain tumors, and (2) identify potential barriers for initial registration of brain tumors in the Alberta Cancer Registry.

METHODS

All patients with a brain tumor diagnosed and residing in Alberta from 2010 to 2015 were extracted, after the DAD review, from the Alberta Cancer Registry (ACR). Descriptive statistics were compiled by behavior and type of registration (originally registered or identified through DAD). The total number of expected nonmalignant brain tumors was estimated by applying the Central Brain Tumor Registry of the United States (CBTRUS) incidence rates to the Alberta population and this estimate was compared to observed numbers. Phi coefficients and χ2 tests for the homogeneity of proportions were conducted to examine bivariate relationships of the characteristics of interest. Multiple logistic regression was used to summarize the independent effects on the probability of being identified through DAD.

RESULTS

The results show 5% of malignant and 35% of nonmalignant brain tumors were identified through DAD review. When comparing observed to expected number of nonmalignant cases after DAD review, the ACR ultimately captured 76% of those expected. Identification through DAD was statistically significantly (P ≤ .05) associated with patients over 75 years old at diagnosis (odds ratio [OR], 2.5), tumors of benign behavior (OR, 2.6), location at diagnosis in Northern Alberta (OR, 1.5), nonmicroscopically confirmed tumors (OR, 1.3), no visit to a CancerControl Alberta facility (OR, 8.7) and certain histological subtypes, including cranial and spinal nerve tumors (OR, 1.7).

CONCLUSION

The use of hospital discharge data significantly improved nonmalignant brain tumor case ascertainment. Therefore, it is recommended that such reviews be instituted annually in provinces while other techniques (such as reminder letters used in Norway or linkages with radiology or other administrative databases) for improving case ascertainment are explored. Those characteristics identified as potential barriers to registration should be investigated to identify possible process improvements in Alberta.

Validation of secondary data sources to identify Parkinson disease against clinical diagnostic criteria

Parkinson disease (PD) is the second most common neurodegenerative disorder. Its diagnosis relies solely on a clinical examination and is not straightforward because no diagnostic test exists. Large, population-based, prospective cohort studies designed to examine other outcomes that are more common than PD might provide cost-efficient alternatives for studying the disease. However, most cohort studies have not implemented rigorous systematic screening for PD. A majority of epidemiologic studies that utilize population-based prospective designs rely on secondary data sources to identify PD cases. Direct validation of these secondary sources against clinical diagnostic criteria is lacking. The Framingham Heart Study has prospectively screened and evaluated participants for PD based on clinical diagnostic criteria. We assessed the predictive value of secondary sources for PD identification relative to clinical diagnostic criteria in the Framingham Heart Study (2001-2012). We found positive predictive values of 1.0 (95% confidence interval: 0.868, 1.0), 1.0 (95% confidence interval: 0.839, 1.0), and 0.50 (95% confidence interval: 0.307, 0.694) for PD identified from self-report, use of antiparkinsonian medications, and Medicare claims, respectively. The negative predictive values were all higher than 0.99. Our results highlight the limitations of using only Medicare claims data and suggest that population-based cohorts may be utilized for the study of PD determined via self-report or medication inventories while preserving a high degree of confidence in the validity of PD case identification.

Use of administrative and electronic health record data for development of automated algorithms for childhood diabetes case ascertainment and type classification: the SEARCH for Diabetes in Youth Study

BACKGROUND

The performance of automated algorithms for childhood diabetes case ascertainment and type classification may differ by demographic characteristics.

OBJECTIVE

This study evaluated the potential of administrative and electronic health record (EHR) data from a large academic care delivery system to conduct diabetes case ascertainment in youth according to type, age, and race/ethnicity.

SUBJECTS

Of 57 767 children aged <20 yr as of 31 December 2011 seen at University of North Carolina Health Care System in 2011 were included.

METHODS

Using an initial algorithm including billing data, patient problem lists, laboratory test results, and diabetes related medications between 1 July 2008 and 31 December 2011, presumptive cases were identified and validated by chart review. More refined algorithms were evaluated by type (type 1 vs. type 2), age (<10 vs. ≥10 yr) and race/ethnicity (non-Hispanic White vs. 'other'). Sensitivity, specificity, and positive predictive value were calculated and compared.

RESULTS

The best algorithm for ascertainment of overall diabetes cases was billing data. The best type 1 algorithm was the ratio of the number of type 1 billing codes to the sum of type 1 and type 2 billing codes ≥0.5. A useful algorithm to ascertain youth with type 2 diabetes with 'other' race/ethnicity was identified. Considerable age and racial/ethnic differences were present in type-non-specific and type 2 algorithms.

CONCLUSIONS

Administrative and EHR data may be used to identify cases of childhood diabetes (any type), and to identify type 1 cases. The performance of type 2 case ascertainment algorithms differed substantially by race/ethnicity.

Are women with major depression in pregnancy identifiable in population health data?

BACKGROUND

Although record linkage of routinely collected health datasets is a valuable research resource, most datasets are established for administrative purposes and not for health outcomes research. In order for meaningful results to be extrapolated to specific populations, the limitations of the data and linkage methodology need to be investigated and clarified. It is the objective of this study to investigate the differences in ascertainment which may arise between a hospital admission dataset and a dispensing claims dataset, using major depression in pregnancy as an example. The safe use of antidepressants in pregnancy is an ongoing issue for clinicians with around 10% of pregnant women suffer from depression. As the birth admission will be the first admission to hospital during their pregnancy for most women, their use of antidepressants, or their depressive condition, may not be revealed to the attending hospital clinicians. This may result in adverse outcomes for the mother and infant.

METHODS

Population-based de-identified data were provided from the Western Australian Data Linkage System linking the administrative health records of women with a delivery to related records from the Midwives' Notification System, the Hospital Morbidity Data System and the national Pharmaceutical Benefits Scheme dataset. The women with depression during their pregnancy were ascertained in two ways: women with dispensing records relating to dispensed antidepressant medicines with an WHO ATC code to the 3rd level, pharmacological subgroup, 'N06A Antidepressants'; and, women with any hospital admission during pregnancy, including the birth admission, if a comorbidity was recorded relating to depression.

RESULTS

From 2002 to 2005, there were 96698 births in WA. At least one antidepressant was dispensed to 4485 (4.6%) pregnant women. There were 3010 (3.1%) women with a comorbidity related to depression recorded on their delivery admission, or other admission to hospital during pregnancy. There were a total of 7495 pregnancies identified by either set of records. Using data linkage, we determined that these records represented 6596 individual pregnancies. Only 899 pregnancies were found in both groups (13.6% of all cases). 80% of women dispensed an antidepressant did not have depression recorded as a comorbidity on their hospital records. A simple capture-recapture calculation suggests the prevalence of depression in this population of pregnant women to be around 16%.

CONCLUSION

No single data source is likely to provide a complete health profile for an individual. For women with depression in pregnancy and dispensed antidepressants, the hospital admission data do not adequately capture all cases.

Estimation of outbreak severity and transmissibility: Influenza A(H1N1)pdm09 in households

BACKGROUND

When an outbreak of a novel pathogen occurs, some of the most pressing questions from a public-health point of view relate to its transmissibility, and the probabilities of different clinical outcomes following infection, to allow an informed response. Estimates of these quantities are often based on household data due to the high potential for transmission in this setting, but typically a rich spectrum of individual-level outcomes (from uninfected to serious illness) are simplified to binary data (infected or not). We address the added benefit from retaining the heterogeneous outcome information in the case of the 2009-10 influenza pandemic, which posed particular problems for estimation of key epidemiological characteristics due to its relatively mild nature and hence low case ascertainment rates.

METHODS

We use mathematical models of within-household transmission and case ascertainment, together with Bayesian statistics to estimate transmission probabilities stratified by household size, the variability of infectiousness of cases, and a set of probabilities describing case ascertainment. This novel approach was applied to data we collected from the early "containment phase" stage of the epidemic in Birmingham, England. We also conducted a comprehensive review of studies of household transmission of influenza A(H1N1)pdm09.

RESULTS

We find large variability in the published estimates of within-household transmissibility of influenza A(H1N1)pdm09 in both model-based studies and those reporting secondary attack rates, finding that these estimates are very sensitive to how an infected case is defined. In particular, we find that reliance on laboratory confirmation alone underestimates the true number of cases, while utilising the heterogeneous range of outcomes (based on case definitions) for household infections allows a far more comprehensive pattern of transmission to be elucidated.

CONCLUSIONS

Differences in household sizes and how cases are defined could account for an appreciable proportion of the reported variability of within-household transmissibility of influenza A(H1N1)pdm09. Retaining and statistically analysing the full spectrum of individual-level outcomes (based on case definitions) rather than taking a potentially arbitrary threshold for infection, provides much-needed additional information. In a future pandemic, our approach could be used as a real-time analysis tool to infer the true number of cases, within-household transmission rates and levels of case ascertainment.

Ability of Medicaid claims data to identify incident cases of breast cancer in the Ohio Medicaid population

BACKGROUND

The use of Medicaid data to study cancer-related outcomes would be highly desirable. However, the accuracy of Medicaid claims data in the identification of incident cases of breast cancer is unknown.

OBJECTIVES

(1) To estimate the sensitivity of Medicaid claims data for case ascertainment of breast cancer, and (2) to determine the positive predictive value (PPV) of diagnostic and procedure codes retrieved from Medicaid claims, using the Ohio Cancer Incidence Surveillance System (OCISS) as the gold standard.

METHODS

The study used the linked OCISS and Medicaid enrollment files, 1997-1998 (n = 1,648). The claims search yielded 2,635 incident cases, of which 1,132 were also identified through the OCISS-Medicaid files. Sensitivity and PPV of Medicaid data were calculated in subgroups of the population.

RESULTS

The overall sensitivity was 68.7 percent, but varied greatly across the subgroups of the population. It was lower among women enrolled in Medicaid only for part of the study year than those enrolled in Medicaid for 12 months of the study year (56.7 percent and 78.0 percent respectively, p < 0.0001), and lower among those who are dual Medicare-Medicaid eligible compared to those not participating in the Medicare program (63.1 percent and 78.6 percent respectively, p < 0.0001). The overall PPV was 43.0 percent, increasing up to 86.6 percent in the presence of procedure codes indicating the presence of mastectomy and lumpectomy, in addition to that of breast cancer diagnosis.

CONCLUSIONS

The sensitivity of Medicaid claims for case ascertainment of breast cancer is somewhat low, but improves considerably when accounting for women enrolled in Medicaid for the entire duration of the study year. The PPV is poor due to a high rate of false positives. The higher PPV obtained in the presence of procedure codes, in addition to diagnosis codes, will help researchers to correctly identify incident cases of breast cancer using Medicaid claims data.