Accuracy of Canadian Health Administrative Databases in Identifying Patients with Rheumatoid Arthritis Using a Random Sample of 7500 Patients Seen in Primary Care

Jessica Widdifield¹, Claire Bombardier², Sasha Bernatsky³, J. Michael Paterson⁴, Jacqueline Young⁴, Diane Green⁴, J. Carter Thorne⁵, Noah Ivers¹, Debra Butt⁴, R. Liisa Jaakkimainen⁶, Myra Wang⁴, Vandana Ahluwalia⁷, George A. Tomlinson⁸ and Karen Tu⁴, ¹University of Toronto, Toronto, ON, Canada, ²Rheumatology, University of Toronto, Toronto, ON, Canada, ³Clinical Epidemiology, Research Institute of the McGill University Health Ctre, Montreal, QC, Canada, ⁴Institute for Clinical Evaluative Sciences, Toronto, ON, Canada, ⁵Southlake Regional Health Centre, Newmarket, ON, Canada, ⁶Preventive Med and Biostatisti, University of Toronto, Toronto, ON, Canada, ⁷William Osler Health Center, Brampton, ON, Canada, ⁸Dept of Medicine/Rheumatology, Toronto General Hospital, Toronto, ON, Canada

Meeting: 2012 ACR/ARHP Annual Meeting

Keywords: diagnosis, epidemiologic methods and rheumatoid arthritis (RA)

Session Information

Title: Epidemiology and Health Services Research: Epidemiology and Outcomes of Rheumatic Disease II

Session Type: Abstract Submissions (ACR)

Background/Purpose:

The use of population-based health administrative databases in rheumatology research is well established, but there are ongoing concerns about validity. To date, previous validation studies have sampled patients primarily from rheumatology clinics, which may limit the usefulness of the results. Our aim was to evaluate the accuracy of administrative data algorithms to identify RA patients drawn from family physician records.

Methods:

We performed a retrospective chart abstraction study using a random sample of 7500 adult patients, age 20 years and over, from the primary care Electronic Medical Record Administrative data Linked Database (EMRALD) in Ontario, Canada. Our reference standard definition for classifying patients as RA included physician-reported RA diagnoses and supporting evidence. RA and non-RA patients were then linked to administrative data to validate different combinations of physician billing (P) and hospitalization (H) diagnostic codes for RA to estimate sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV).

Results:

Based on our reference standard definition, we identified 69 patients with physician-reported RA for an overall RA prevalence of 0.92%. Most RA cases were female (64%) and the mean (SD) age was 62 (14) years. Among RA cases, 86% had a documented diagnosis by a specialist and 80% had documentation of a disease-modifying anti-rheumatic drug exposure. Test characteristics of selected RA case definition algorithms tested are reported in Table 1. All algorithms tested had excellent specificity (97-100%), however sensitivity varied (75-90%) among physician billing diagnosis algorithms. Despite the low RA prevalence, algorithms for identifying RA patients had modest PPV, which improved substantially with the requirement of having musculoskeletal specialist billing codes for RA (51-83%). Varying the observation window had little impact on the accuracy of the algorithms tested.

Conclusion:

The RA case definition algorithms that we tested had excellent specificity. To our knowledge, this is the first study to rigorously evaluate the accuracy of RA administrative data algorithms in a random sample from family physician records. We are independently validating these algorithms in a random sample of patients from rheumatology clinics to support the findings of this work.

Table 1: Test characteristics of selected algorithms
Algorithm	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)
1 H ever	22	100	88	99
1 P ever	90	97	20	100
2 P in 1 year	84	99	46	100
2 P in 2 years	84	99	45	100
2 P in 3 years	84	100	42	100
3 P in 1 year	80	100	63	100
3 P in 2 years	80	100	60	100
3 P in 3 years	80	100	59	100
1 P ever by a specialist	81	99	51	100
2 P in 1 year at least 1 P by a specialist	78	100	65	100
2 P in 2 years at least 1 P by a specialist	78	100	65	100
2 P in 3 years at least 1 P by a specialist	78	100	62	100
3 P in 1 year at least 2 P by a specialist	75	100	83	100
3 P in 2 years at least 2 P by a specialist	75	100	81	100
3 P in 3 years at least 2 P by a specialist	75	100	81	100
1 H or 3 P in 1 year at least 1 P by a specialist	78	100	77	100
1 H or 3 P in 2 years at least 1 P by a specialist	78	100	76	100
1 H or 3 P in 3 years at least 1 P by a specialist	78	100	76	100
H: Hospitalization code; P=physician diagnostic code; Specialist = rheumatologist, internal medicine, orthopedic surgeon

Disclosure:

J. Widdifield,
None;

C. Bombardier,
None;

S. Bernatsky,
None;

J. M. Paterson,
None;

J. Young,
None;

D. Green,
None;

J. C. Thorne,
None;

N. Ivers,
None;

D. Butt,
None;

R. L. Jaakkimainen,
None;

M. Wang,
None;

V. Ahluwalia,
None;

G. A. Tomlinson,
None;

K. Tu,
None.

« Back to 2012 ACR/ARHP Annual Meeting

ACR Meeting Abstracts - https://acrabstracts.org/abstract/accuracy-of-canadian-health-administrative-databases-in-identifying-patients-with-rheumatoid-arthritis-using-a-random-sample-of-7500-patients-seen-in-primary-care/