Validation of a Sampling Approach to Estimate Lyme Disease Cases in a State in Which the Etiologic Agent is Endemic in Every County

BACKGROUND: Lyme disease case investigation in endemic states is a significant burden. One approach to reducing the burden is application of a random case sampling strategy, rather than follow-up of every case, to produce estimated case counts. New York pilot tested and currently utilizes a 1:5 sampling ratio. Initial results from Minnesota’s analysis of 2010 data evaluating 1:2 and 1:5 case sampling ratios indicate no statistically significant differences between sampled and unsampled groups. New York and Minnesota are Lyme endemic states, but both have non-endemic counties. This study utilized the Minnesota algorithm on data from Massachusetts, where Lyme disease is endemic in every county.

METHODS: Separate random samples of reports initiated via laboratory reporting were generated using 1:2 and 1:5 sampling ratios for 2010 and 2011 data. Estimates of confirmed case counts were calculated by multiplying the number of confirmed reports in the sample by either 2 or 5 and adding the number of confirmed reports initiated from medical providers. Sample groups were compared to cases using Chi-square goodness of fit tests to assess statistical differences in the proportions of erythema migrans (EM), were male, were pediatric (≤ 12 years of age), or had late manifestations of disease. Comparison of the distribution of cases by county of residence was performed similarly.

RESULTS: 2,670 actual confirmed cases of Lyme disease were reported in 2010; the 1:2 and 1:5 sampling strategies produced estimates of 2,647 and 2,790 respectively. 2,814 confirmed cases in 2011 produced estimates of 2,805 and 2,894. There were no significant differences in proportions of EM, male gender, pediatric age or late manifestations in the 1:2 sample for either 2010 or 2011 (p>0.05). There was a statistically significant difference in the proportion of pediatric cases using the 1:5 sampling ratio. This was seen in both 2010 (p=0.0036) and 2011 (p=0.0364). The case distribution across counties of residence was not significantly different by sampling ratio or year.

CONCLUSIONS: This analysis confirms that estimating Lyme disease case counts using either a 1:2 or a 1:5 sampling strategy produces results that are not significantly different from the complete results on several major parameters. Underrepresentation of young children using the 1:5 sampling ratio may be explained by the fact that Lyme disease reports on pediatric cases are more likely to originate from providers, not laboratories. Therefore, cases in children are under-represented in the sampled laboratory reporting data and the smaller the sampling ratio, the greater the resulting discrepancy.