Under-detection of blood culture-positive enteric fever cases: The impact of missing data and methods for adjusting incidence estimates

Abstract

Background: In surveillance for typhoid fever, under-detection of cases occurs when patients with fever do not seek medical care, or seek medical care but do not receive a blood test. Missing data may result in incorrect estimates of disease incidence.

Methods: We used data from an ongoing randomised clinical trial of typhoid conjugate vaccine among children in Nepal to determine if eligible patients attending our fever clinics who did not have blood taken for culture had a lower risk of disease than those who had blood drawn. We assessed clinical and demographic predictors of having blood taken for culture, and predictors of culture-positive results. Missing blood culture data were imputed using multiple imputations.

Results: During the first year of surveillance, 2392 fever presentations were recorded and 1615 (68%) of these had blood cultures. Children were more likely to have blood taken for culture if they were older, had fever for longer, a current temperature ≥38 degrees, or if typhoid or a urinary tract infection were suspected. Based on imputation models, those with blood cultures were 1.87 times more likely to have blood culture-positive fever than those with missing data.

Conclusion: Clinical opinion on the cause of the fever may play a large part in the decision to offer blood culture, regardless of study protocol. Crude typhoid incidence estimates should be adjusted for the proportion of cases that go undetected due to missing blood cultures while adjusting for the lower likelihood of culture-positivity in the group with missing data.

Fig 1

Proportion of trial participants who have blood drawn for culture (A and B) or are blood culture positive (C and D) when presenting to fever clinics with either 2 days of self-reported fever or a current temperature of 38 degrees. Proportion of children presenting to fever clinics who have blood drawn for culture by (A) age and clinical suspicion of typhoid (both p<0.0001), (B) fever duration (p<0.0001) and current temperature (p = 0.007). Proportion of children who have blood culture-positive enteric fever by (C) age and clinical suspicion of typhoid, (D) days with fever and current temperature. Circles represent the proportion with blood taken for each 1-year age band (A and C), or in groups according to number of days of fever (B and D). The size of the circle is proportional to the number of fever presentations in that group.

Fig 2. Results of 100 multiple imputation models imputing blood culture results for eligible participants presenting to fever clinics who did not have a blood culture taken.

(A) Each dot shows the percentage of patients with culture-positive results in one imputed dataset, i.e. those with no blood drawn for culture (blue). The dashed blue line shows the median percent culture-positive in the imputed data, whereas the dashed red lines shows the observed percent culture-positive in those who had a blood culture taken. (B) The proportion culture-positive (observed data: red, imputed data: blue) is plotted along with the smooth spline showing that the observed data underestimate culture-positive rates in very young children less than 4 years of age and overestimate culture-positive rates in older children. Generalised additive models (GAMs) were used derive smoothed line of best fit and confidence intervals. GAMs for binary data with logit-link functions were implemented using cubic spline penalized regression with automatic smoothness selection.