Investigating biases in routine pharmaceutical data collections: An evaluation of the national pharmaceutical data collection for assessing medicine adherence in New Zealand
Many medicines require patients to take them continuously to be effective. Routine data collections provide a possible data source for assessing medication adherence at the population level, but can be limited by biases in their information capture. We investigated the impact of such biases by comparing levels of suboptimal adherence to allopurinol (used for gout control) across population groups using routine data and data obtained directly from pharmacies.
METHODS:
Allopurinol dispensing records from the Gisborne area (population: 44,463) of New Zealand were obtained for October 2005 – September 2006 from both the national pharmaceuticals dispensing data collection and community pharmacies. Data on age, sex and ethnicity were obtained from the national healthcare user collection. Adherence was assessed using the Medicine Possession Ratio (MPR), the proportion of days in a defined period that a person possesses a medicine. Odds ratios for suboptimal adherence (MPR < 0.80) when using routine and direct pharmacy data were calculated using logistic regression.
RESULTS:
Fewer (615 versus 805, 24% fewer) allopurinol users were identified using routine data compared to pharmacy data. This varied by population group, with 30% fewer Māori (the first people of New Zealand) identified using routine data. The pattern of odds ratios across population groups did not differ, with Māori more likely to have suboptimal adherence compared to non- Māori (OR: 1.52 (95% CI: 1.01 – 2.29)) versus 2.51 (95% CI: 1.78 – 3.54)). Odds ratios were attenuated when using routine data.
CONCLUSIONS:
While using routine data did not alter the pattern of suboptimal adherence, it did attenuate the estimates. It also differentially affected absolute estimates of allopurinol use and suboptimal adherence across groups. These findings are directly relevant to the New Zealand context but, as the use of routine data collections increases, they also highlight the importance of assessing potential biases in these collections.