Elsevier

Vaccine

Volume 31, Issue 30, 26 June 2013, Pages 3104-3109
Vaccine

The case test-negative design for studies of the effectiveness of influenza vaccine

https://doi.org/10.1016/j.vaccine.2013.04.026Get rights and content

Highlights

  • We examine properties of the “case test-negative” design.

  • Vaccine effectiveness estimates from such studies appear valid under general assumptions.

  • If influenza and non-influenza infections differ in disease severity then effectiveness estimates may be biased.

  • If acute respiratory viral infections interact via transient non-specific immunity, vaccine effectiveness estimates from case test-negative studies are biased, but the bias will be non-trivial only if influenza incidence is extremely high and the duration of the transient non-specific immunity is long.

Abstract

Background

A modification to the case–control study design has become popular to assess vaccine effectiveness (VE) against viral infections. Subjects with symptomatic illness seeking medical care are tested by a highly specific polymerase chain reaction (PCR) assay for the detection of the infection of interest. Cases are subjects testing positive for the virus; those testing negative represent the comparison group. Influenza and rotavirus VE studies using this design are often termed “test-negative case-control” studies, but this design has not been formally described or evaluated. We explicitly state several assumptions of the design and examine the conditions under which VE estimates derived with it are valid and unbiased.

Methods

We derived mathematical expressions for VE estimators obtained using this design and examined their statistical properties. We used simulation methods to test the validity of the estimators and illustrate their performance using an influenza VE study as an example.

Results

Because the marginal ratio of cases to non-cases is unknown during enrollment, this design is not a traditional case-control study; we suggest the name “case test-negative” design. Under sets of increasingly general assumptions, we found that the case test-negative design can provide unbiased VE estimates. However, differences in health care-seeking behavior among cases and non-cases by vaccine status, strong viral interference, or modification of the probability of symptomatic illness by vaccine status can bias VE estimates.

Conclusions

Vaccine effectiveness estimates derived from case test-negative studies are valid and unbiased under a wide range of assumptions. However, if vaccinated cases are less severely ill and seek care less frequently than unvaccinated cases, then an appropriate adjustment for illness severity is required to avoid bias in effectiveness estimates. Viral interference will lead to a non-trivial bias in the vaccine effectiveness estimate from case test-negative studies only when incidence of influenza is extremely high and duration of transient non-specific immunity is long.

Introduction

A modification of the traditional case–control study design often termed the “test-negative case–control” design has become popular for post-licensure observational studies of the effectiveness of vaccines for influenza [1], [2], [3], [4], [5], [6], [7], [8], [9] and rotavirus [10], [11], [12], [13], [14], [15]. In this design, patients seeking medical care for a defined clinical condition (e.g., acute respiratory illness) are tested for a specific viral infection (e.g., influenza) by using a highly sensitive and specific laboratory test, usually a polymerase chain reaction (PCR) assay. Those testing positive are cases. Controls are patients meeting the same enrollment criteria but who test negative for infection. Vaccine effectiveness (VE) is calculated in the usual manner for case-control studies, i.e., VE = (1  vaccination odds ratio) x 100%. As the marginal ratio of cases to non-cases or “controls” is not specified or even knowable during enrollment, which occurs prior to testing, this design is clearly not a traditional case–control study. For clarity, we will refer to this design as a “case test-negative” design hereafter. The recent popularity of this design arises primarily from ease of implementation, since both cases and controls are recruited in one process. Another advantage is the likelihood that enrollment in this manner reduces the risk of particular type confounding: if vaccine receipt is associated with a greater likelihood of seeking health care for mild to moderate illnesses, then the design should adjust implicitly for this confounding, which otherwise would bias VE estimates downwards [3], [9]. Although the case test-negative approach is appealing both intuitively and practically, its assumptions have not been stated explicitly nor have the statistical properties of the VE estimator been derived formally. Using an observational study of influenza VE as an example, we derive the mathematical foundation of the case test-negative design and determine the conditions under which VE estimates derived with it are valid and unbiased.

Section snippets

Methods

We examined the properties of the case test-negative design in two ways. First, we derived mathematical expressions for VE estimates, starting with the most constraining assumptions and then relaxing them to address a broader range of situations. Second, we used simulation methods to examine the validity of the VE estimates and to illustrat their performance.

Results

Based on a derivation of the parameters obtained from a hypothetical case test-negative study of the effectiveness of influenza vaccination, we found that under our base case assumptions the VE estimate was valid and unbiased (Eq. (2)). The estimate was also valid in the context of time-varying incidence rates (Eq. (A.3), Appendix 1). However, if health care-seeking behavior is not only driven by the attributes that also drive vaccination uptake, but also by disease severity and if disease

Discussion

We found that, under increasingly general assumptions, estimates for the effectiveness of vaccine against symptomatic influenza, obtained from case test-negative studies, are valid and unbiased. Thus, standard statistical methods can be used to analyze data from such studies, and VE can be calculated directly from the estimated exposure odds ratio, VE = (1  odds ratio) x 100%, as done when analyzing case–control study data.

Our derivations and the simulations demonstrated that the case test-negative

Acknowledgements

The authors wish to thank Michael Jackson, Paul Gargiullo and Sue Reynolds for their helpful comments.

References (22)

  • J.A. Boom et al.

    Effectiveness of pentavalent rotavirus vaccine in a large urban population in the United States

    Pediatrics

    (2010)
  • Cited by (0)

    The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the Centers for Disease Control and Prevention.

    View full text