Validation of administrative data case definitions for gestational diabetes mellitus.

AIM

To examine, using administrative data, the validity of two algorithms for identifying gestational diabetes mellitus: 1) the current National Diabetes Surveillance System algorithm for excluding gestational diabetes cases and 2) gestational diabetes-specific ICD codes in the delivery-related hospitalization.

METHODS

This was a retrospective study of all women, aged 18-54 years, residing in Alberta, Canada, with singleton deliveries between 1 April 1999 and 31 March 2010. We linked Alberta Perinatal Health Program data on all deliveries to administrative claims data from Alberta Health using the mother's personal health number. For both gestational diabetes algorithms, we calculated the sensitivity, specificity, positive predictive value, negative predictive value and agreement, using gestational diabetes identified in the Alberta Perinatal Health Program as the 'gold standard'.

RESULTS

Our study sample consisted of 411 390 deliveries for 273 152 women. The mean (sd) age was 29.1 (5.6) years and 82.3% of the women were white. Crude rates of gestational diabetes were 3.9% (16 215 cases), 1.3% (5189 cases) and 4.0% (16 440 cases) according to the Alberta Perinatal Health Program, National Diabetes Surveillance System and ICD code-based algorithms, respectively. Compared with the Alberta Perinatal Health Program database, the National Diabetes Surveillance System algorithm had a sensitivity of 25% and specificity of 100%, whereas the gestational diabetes-specific ICD code-based algorithm had a sensitivity of 86% and specificity of 99%.

CONCLUSIONS

The National Diabetes Surveillance System algorithm underestimates the number of gestational diabetes cases. A more valid mechanism to identify gestational diabetes prevalence using health administrative data is the use of gestational diabetes-specific ICD-9/10 codes in the delivery hospitalization.