Modelling diagnostic error to improve estimation of helminth treatment efficacy, age-specific prevalence and enhance comparison of diagnostic tools

Background: According to the latest global disease burden estimates roughly 1.1 billion people suffer from neglected tropical diseases (NTDs), from which over a billion are infected with soil-transmitted helminths (STHs) or Schistosoma leading to a considerable public health burden. For instance, children mainly suffer from anaemia and growth stunting while adults may develop chronic diseases, which can also result in confined function of affected organs. With financial support from different organisations and donations from pharmaceutical companies, numerous interventions have been implemented to reduce the burden of NTDs. In some cases, specific NTDs have been eliminated. The main interventions are preventive chemotherapy (PC), water, sanitation and hygiene (WASH) and information, education and communication (IEC). However, much remains to be done, and hence, in 2020, the World Health Organization (WHO) put forth a road map for the control and elimination of the NTDs.
To reach the best possible outcomes, control programmes need to be informed about the impact of interventions. This requires prevalence and infection intensity estimates at baseline and following interventions. To assess the prevalence and efficacy of treatments, tools that are widely used lack accuracy, and hence, result in underestimated prevalences and overestimated cure rates (CRs). Highly sensitive diagnostic techniques are available, but rarely applied in the field. It is therefore important to include the diagnostic error to obtain ‘true’ estimates to accurately inform disease control programmes.
Goal and objectives: The overreaching goal of this PhD thesis was to improve estimation of diagnostic error taking into account variation in egg counts from day- to-day and slide-to-slide to evaluate the performance of diagnostics, disease burden and the efficacy of treatments to better guide control and elimination efforts of NTDs. The specific objectives were to (i) estimate the performance of diagnostic tools against Schistosoma haematobium; (ii) translate prevalence thresholds from urine filtration into reagent strip; (iii) assess the efficacy of drug therapies against STHs; (iv) assess the sensitivity of diagnostics for different Schistosoma species and STHs; and (v) estimate age-dependent prevalence for S. mansoni using Bayesian methods.
Methods: In Chapter 2, I developed a model to compare diagnostic methods for S. haematobium, namely urine filtration and reagent strip. Urine filtration results consisted of egg counts, while reagent strip results were semi-quantitative, wherefore two different distributions were assumed. To estimate the prevalence, the infected and non-infected individuals were separated. As the data consist of specimen taken on five consecutive days subjected to two different tests, I was able to take into account diagnostic error and hence, estimate infection intensity-dependent sensitivity. Moreover, by running extensive simulations of hypothetical populations in diverse transmission settings, we relate WHO prevalence thresholds from urine filtration into reagent strip.
In Chapter 3, I developed a transmission model to conduct a meta-analysis on CR and egg reduction rate (ERR) for an ensemble of drug therapies against hook- worm. The prevalence, which is defined as harbouring at least one fertilized female worm, and the CRs were estimated by incorporating a mixture modelling approach. The model was fitted to data from six randomized controlled trials. At baseline and treatment follow-up, two specimen per study participant were collected over consec- utive days, which enabled us to take into account diagnostic error and estimate the infection intensity-dependent sensitivity of the Kato-Katz test for hookworm.
In Chapter 4, I extended the egg count model of Chapter 3 and included the density-dependent fecundity to estimate CR and ERR for several drug therapies against Trichuris trichiura. Moreover, the infection intensity-dependent sensitivity of the Kato-Katz test was estimated for T. trichiura.
In Chapter 5, I extended a transmission model and included diagnostic error to estimate age-dependent prevalence curves for S. mansoni and predicted the preva- lence of adults from the prevalence of school-aged children. The model was fitted to data from Uganda, which was collected by the Schistosomiasis Consortium for Operational Research and Evaluation (SCORE) and consists of specimen taken on consecutive days (two days for two years, three days for one year). This enabled estimating the infection intensity-dependent sensitivity of the Kato-Katz test for S. mansoni.
Results: The diagnostic accuracy of reagent strip was equivalent to urine filtra- tion data obtained on a single day, when traces were considered negative. A 10% and 50% urine filtration prevalence based on a single day sampling corresponds to 11.2% and 48.6% prevalence by reagent strip, respectively, when traces were con- sidered negative, and 17.6% and 57.7%, respectively, when traces were considered positive.
Taking the diagnostic error into account resulted in considerably lower CRs for drug therapies against hookworm than previously reported. Overall, of all treat- ments analyzed, mebendazole administered in six dosages of 100 mg each was the most efficacious treatment with a CR of 88% (95% Bayesian credible interval: 79- 95%). Diagnostic sensitivity of Kato-Katz for hookworm varied with the infection intensity and sampling effort. For an infection intensity of 50 eggs per gram of stool (EPG), the sensitivity is close to 60%; for two Kato-Katz thick smears it increased to 76%.
The treatment with the highest CR against T. trichiura was the combination therapy of albendazole plus pyrantel pamoate plus oxantel pamoate with a CR of 79% and an ERR of 91%. Albendazole plus oxantel pamoate showed the highest ERR of 97% and a CR of 69%. For 24 EPG, the sensitivity was around 50% for a single and increased to almost 70% for duplicate Kato-Katz thick smears.
For 24 EPG, the sensitivity of Kato-Katz for S.mansoni was estimated to be 55%, 77% and 99% for simple, duplicate and quadruplicate thick smears.
Conclusions/significance: The main contribution of this PhD thesis to the field of mathematical modelling in public health is the development or extension of modelling frameworks to accurately compare the performance of diagnostics, assess their sensitivity, estimate treatment efficacy and age-specific prevalence. In more detail: (i) reagent strip and urine filtration were compared for detection of S. haematobium; (ii) the intensity-dependent sensitivity for different sampling schemes was estimated for aforementioned diagnostic tools as well as the Kato-Katz technique for hookworm and T. trichiura; (iii) the efficacy of different treatments against hook- worm and T. trichiura were assessed; (iv) the age-specific prevalence for S. mansoni was estimated; and the prevalence in adults was predicted from the prevalence in school-aged children for S. mansoni. The higher accuracy compared to existing studies was achieved by taking into account diagnostic error and the transmission mechanism.
If reagent strip instead of urine filtration would be employed in schistosomiasis control programmes, the costs could be substantially reduced and proceedings would be more efficient. Moreover, the treatment efficacy and infection intensity- dependent sensitivity estimates for different sampling schemes may be considered in aforementioned programmes to make decisions about suitable drug therapies and sampling designs.