Modeling Epidemiologic Typing Data and Likelihood Inference for Disease Spread

Article excerpt

A model for epidemiologic typing data is introduced, and likelihood ratio methods are developed for evaluating these data as evidence about disease spread. The observed data consist of microorganism subtypes from an index case of infectious disease, cases clustered with the index case, and a reference sample. The likelihood methods are evaluated via probabilities of observing epidemiologic subtypes that represent strong and sometimes misleading evidence favoring one hypothesis over another. A general bound is identified for the probability of observing strong evidence favoring a close epidemiologic relationship between the index and cluster cases vis-a-vis no relationship when in fact there is none. The advantages of this approach versus alternate approaches to measuring the strength of typing evidence are discussed.

KEY WORDS: Likelihood ratio; Misleading evidence; Molecular epidemiology; Profile likelihood; Reference sample; Schwarz criterion.

1. INTRODUCTION

1.1 Background

Outbreaks of infectious disease in human populations occur through the transmission of disease-causing microorganisms called pathogens. A major goal of outbreak investigations is to prevent further transmission of disease. The control of transmission is most effective when the pattern of transmission among the observed cases of disease is known. The disease transmission pattern is typically investigated with the aid of molecular methods (Maslow, Mulligan, and Arbeit 1993). These methods rely on the fact that for most infectious diseases, nearly all of the culprit organism's genetic characteristics remain constant during infectious cycles in hosts and during transmission. The molecular methods are thus used to identify subtypes of microorganisms among those isolated from clustered cases of disease that might or might not constitute an outbreak. The characteristics measured in subtyping procedures vary from method to method, but regardless of the method used, the aim is to delineate outbreak-related and -unrel ated (i.e., sporadic) cases of infectious disease. The results are then used together with results from other epidemiologic methods, such as the identification of personal contacts, to assess possible patterns of transmission.

To illustrate how molecular typing data might be used, consider the following scenario. Suppose that a worker in a health care facility becomes ill and is subsequently diagnosed with an infectious disease known to be spread through person-to-person contact. The possibility that he or she acquired the infection from a patient is considered. A review of records from recent patients in the facility identifies a greater than expected number of patients with the same infection. Included is a patient who had contact with the given health care worker. This worker is then the first identified case, called the index case, in this possible outbreak of infectious disease.

To initiate the subtyping process, blood samples are obtained for culturing microorganisms from the index case and all known concurrent cases of the particular infectious disease in the facility. Cultured microorganisms from possibly related previous cases are also retrieved when available from laboratory storage. These microorganisms are then subtyped. Suppose that the subtypes from the index case and his or her patient contact are the same. Furthermore, suppose that nearly all of the remaining microorganisms have this subtype as well. Then it is possible that cases with this particular subtype are related through transmission--that is, they are epidemiologically related--whereas cases with different subtypes are probably unrelated.

It is possible, however, that the observed subtype from the index and other cases is simply a widespread subtype for this pathogen. The subtyping procedure is thus carried out on a control group of patients who had the same infectious disease and were cared for in the same facility during the previous year, but whose infection is believed to be unrelated to the cases already studied and to one other. …