Comparison of two alternative study designs in assessment of medicines utilisation in neonates

BACKGROUND: Estimates of prevalence are known to be affected by the design of cross-sectional studies. A pan-European study provided an opportunity to compare the effect of two cross-sectional study designs on estimates of medicines use. METHODS: A Service evaluation survey (SES) and a web-based point-prevalence study (PPS) were conducted as part of a European study of neonatal exposure to excipients. Neonatal units from all European Union countries plus Iceland, Norway, Switzerland and Serbia were invited to participate. All medicines prescribed to neonates were recorded during three-day and one-day study periods in the SES and PPS, respectively. In the PPS individual demographic and prescription data were also collected. To compare the probabilities that a particular medicine would be reported by each study multilevel mixed effects logistic regression models with crossed random effects were applied. The relationship between medicines exposure at the unit and individual levels in the PPS data was assessed using polynomial regression with square root transformation. RESULTS: Of 31 invited countries 20 and 21 with 115 and 89 units joined the SES and PPS, respectively. Out of 5,572,859 live births in invited countries in 2010 a higher proportion was covered by units participating in the SES compared to the PPS (11% vs 6%, respectively; OR 1.89; 95% CI 1.87-1.89). A greater number of active pharmaceutical ingredients (API), manufacturers and trade names were registered in the SES compared to the PPS. High correlation between the two studies in frequency of use for each specified API was seen (R(2) = 0.86). The average probability of a department to use a given API was greater in the SES compared to the PPS (OR 2.36; 95% CI 2.05-2.73) with higher frequency of use and longer average duration of prescription further increasing the difference. The polynomial regression model described the correlation between APIs exposure on unit and individual level well (R(2) = 0.93). CONCLUSION: The simple data structure and longer study period of the SES resulted in improved recruitment and higher likelihood of capture for a given API. The frequency of use at the unit level appears a good surrogate of individual exposure rates.