29 Aug Identification of incident pancreatic cancer in Ontario administrative health data
Identification of incident pancreatic cancer in Ontario administrative health data
What is the aim of the study?
To validate three approaches for identifying incident cases of pancreatic cancer in Ontario administrative claims data.
How was the study conducted?
- We created a cohort using Ontario (Canada) administrative health data from 2002 to 2012 and identified cases of pancreatic cancer with three approaches, using the Ontario Cancer Registry (OCR) as the reference standard.
- In the any diagnosis approach, cases were defined by primary or secondary diagnostic codes for pancreatic cancer in outpatient or inpatient records. In the any inpatient diagnosis approach, cases were defined using only diagnoses in hospital discharge abstracts. In the algorithm approach, cases were identified by an algorithm that combined the first two approaches.
- Comparing each approach to the OCR, we calculated the expected value and 95% confidence interval (CI) of the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV).
- We also compared the event dates using each approach with those recorded in the OCR.
What did the study find?
- Among a total of 12 060 837 patients in Ontario administrative health data sources, 13 999 incident pancreatic cancer cases were identified in the OCR. Sensitivity ranged from 72.5% (algorithm) to 97.5% (any diagnosis), and PPV ranged from 38.4% (any diagnosis) to 78.9% (any inpatient diagnosis). Specificity and NPV were ~100% for all approaches. The median absolute difference in cancer event date ranged 0 to 15 days.
- The any inpatient diagnosis method had the highest PPV (78.9%; 95% CI: 78.2‐79.5%) and moderate sensitivity (86.6%; 95% CI: 86.0‐ 87.2%).
Inpatient diagnoses of pancreatic cancer in Ontario administrative heath data are suitable for pancreatic cancer case identification.
- Few validation studies have evaluated incident pancreatic cancer case definitions in administrative health databases compared with cancer registries, considered the reference standard.
- Three approaches for identifying pancreatic cancer cases were evaluated using (1) inpatient diagnoses only; (2) any inpatient or outpatient diagnosis; and (3) an algorithm that combined 1 and 2.
- The optimal approach was to use inpatient diagnoses only, which yielded the highest positive predictive value, had moderate sensitivity, and did not appreciably affect the date of cancer diagnosis compared with that recorded in the cancer registry (median difference = 0 days).
- In the absence of cancer registry data, future epidemiologic studies could use inpatient diagnosis codes to identify incident pancreatic cancer.
Wu JW, Azoulay L, Huang A, Paterson M, Wu F, Secrest MH, Filion KB. Identification of incident pancreatic cancer in Ontario administrative health data: A validation study. Pharmacoepidemiol Drug Saf. 2020 Jan;29 Suppl 1:78-85.