Analysis of the impact of misclassification bias on hospital surveillance results

Infection control practitioners (ICPs) and epidemiologists often rely on hospital surveillance results to guide outbreak investigations and to evaluate the effectiveness of infection prevention and control practices. Unfortunately, surveillance results are highly vulnerable to systematic error due to misclassification bias because of the large numbers of health care providers involved in the data collection process, and because of low incidents of infection. Quantitative assessments of misclassification bias can demonstrate its impact on surveillance results and can help decide whether it is a plausible explanation for surveillance findings. The purpose of this paper is to provide an in-depth analysis of the effect of misclassification bias on data obtained from hospital surveillance. In order to do this, we used the 2004 Surrey Memorial Hospital Caesarian section outbreak investigation and a fictitious study of ventilator-associated pneumonia (VAP). Statistically insignificant variability between two groups of ICPs was shown to have important clinical significance because of major effects on the incidence of infection calculated. Valid and useful surveillance results were shown to only be obtainable when a single case definition with a high positive predic- tive value and a very high degree of clarity, sensitivity and specificity was consistently used by all health care professionals involved in the surveil- lance data collection process, and when surveillance was targeted to high prevalence infections. The Centers for Disease Control and Prevention (CDC) defines surveillance as the “ongoing systematic collection, analysis, and interpretation of health data essential to the planning, implementation, and evalua- tion of public health practice closely integrated with the timely dissemination of these data to those who need to know”. It is the method used by infection control practitioners and public health officers to detect fluc- tuations in dangerous or indicator infections within the hospital envi- ronment and in the community. The data gathered can be used to compare current rates to previous rates within the same environment or population, and to compare with regional, national or international rates. Surveillance findings can therefore be powerful tools for outbreak investigations, to evaluate control measures, for quality improvement and for supporting prac- tice and policy changes at all levels within the health care system. Nevertheless, the data collected using surveillance are only useful if they are valid and reliable. Statistical methods of analysis often only account for random error and control- lable confounding, and only explain a small portion of the total error that can affect surveillance results. A large portion of the total error results from the measures used to seek out cases, as well as from the frequency of misclassification of cases and non-cases. The purpose of this paper is to provide an in-depth analysis of the effect of misclassification bias on the data obtained from hospital surveillance. Two surveillance scenarios will be used to illustrate why misclassification may occur and how misclassification bias may affect surveillance findings. Actual data from a recent outbreak of surgical site infections (SSI) was used for the first scenario, and data for the second senario was derived from literature on ventilator-associated pneumonia (VAP).