Association Between Malaria and Invasive Nontyphoidal Salmonella Infection in a Hospital Study: Accounting for Berkson's Bias

Background: There is growing evidence for a positive association between malaria and invasive nontyphoidal Salmonella (iNTS) disease. However, case-control studies conducted within healthcare facilities also report inverse associations. This may be due to Berkson's bias, a selection bias that acts when both exposure and outcome are associated with hospital attendance and study participants are selected among attendees only. This study describes the effect of Berkson's bias on the malaria-iNTS association and provides a less biased effect estimate.

Methods: Data collected in 2 Ghanaian hospitals were analyzed using 2 case-control approaches. In both approaches, cases were defined as iNTS-positive children, and concomitant malaria infection was the exposure of interest. In the first conventional sampling approach, children without any febrile bloodstream infection served as controls. In the second control-disease approach, children with non-iNTS bacteremia were used as controls.

Results: Data from 6746 children were suitable for the analyses. One hundred sixty children with iNTS infection were study cases. In the conventional case-control approach 6301 children were controls, and in the control-disease approach 285 children were controls. In the conventional case-control study, malaria was estimated to protect against iNTS disease (odds ratio [OR], 0.4; 95% confidence interval [CI], .3-.7), whereas in the control-disease approach, malaria was identified to be a risk factor for iNTS disease (OR, 1.9; 95% CI, 1.1-3.3).

Conclusions: The study highlights how a selection bias may reverse results if an unsuitable control group is used and adds further evidence on the malaria-iNTS disease association.