Using administrative health data to ascertain ovarian cancer treatment information in Australia: a validation against medical records

OBJECTIVES: Understanding patterns of cancer care is important for improving patient safety and reducing inequity. Linked administrative datasets offer a cost-efficient alternative to medical chart abstraction, but the information may not be as comprehensive and accurate. Our aim was to determine the validity of using linked administrative data to identify ovarian cancer primary treatment in Australia, using data from medical charts for comparison. METHODS: We used data from the Ovarian cancer Prognosis And Lifestyle study, a national prospective cohort study including women diagnosed with ovarian cancer from 2012 to 2015. Treatment information collected via medical records was used as the reference standard. Linked administrative data were obtained from national (Medicare Benefits Schedule, Pharmaceutical Benefits Scheme) and state/territory (hospital admitted patient data collections) data linkage units. The data were partitioned into training (60%, n = 413) and validation (40%, n = 271) datasets. Initial definitions were developed and further refined using the training set and based on clinical advice. The final definitions were assessed using the validation set. Sensitivity, specificity, negative predictive value, positive predictive value and kappa were calculated for receipt of treatment, whereas agreement was assessed for dates and continuous measures. RESULTS: Sensitivity and specificity were high for receipt of chemotherapy (99%, 100%), cytoreductive surgery (97%, 100%), platinum-based chemotherapy (97%, 100%), taxane-based chemotherapy (98%, 88%), intra-peritoneal chemotherapy (93%, 99%) and whether chemotherapy was given in the adjuvant setting (100%, 90%). Agreement for treatment dates was high for surgery (exact agreement 90%, 97% ±7 days) and first chemotherapy (85%, 96% ±7 days). Agreement was moderately high for end of primary chemotherapy (72%, 82% ±7 days), number of chemotherapy cycles (65%, 89% ±1 cycle) and chemotherapy regimen (86%). CONCLUSIONS: Linked administrative data can provide accurate information on core aspects of ovarian cancer treatment when there are barriers to accessing patient medical records.