Search

Background: With 1.4 million deaths in 2019,tuberculosis remains a leading global infectious cause of death, second only to COVID-19 in 2020. Antimicrobial resistant infections are projected to cause over 10 million deaths annually by 2050,in part reflecting non-pathogen directed prescription secondary to limited point-of-care diagnostics and laboratory infrastructure. These two major global public health threats intersect through diagnostic algorithms that have for decades encouraged broad-spectrum antibiotic prescriptions("trial-of-antibiotics") during the diagnostic work-up leading to ~5 million people being treated annually for mycobacteriology-negative tuberculosis. However, the underlying assumption that post-treatment symptom improvement "rules out" tuberculosis, had no clear evidence-base. Aims: To use systematic review and a randomised controlled trial to evaluate and address evidence gaps in three key areas:1) diagnostic performance, 2) safety of withheld prescription, and 3) impact on antimicrobial resistance of trial-of-antibiotics. Methods and results: For the systematic review and meta-analysis (CRD42017083915), I searched MEDLINE, Embase and Global Health databases for studies addressing the diagnostic performance of trial-of-antibiotics against mycobacteriology tests in adults with tuberculosis symptoms. Pooled values for sensitivity and specificity of trial-of-antibiotics (index test) versus mycobacteriology tests (reference) were estimated using random-effects bivariate modelling and I2statistic. Only 8/9410 screened studies were eligible, with no randomised trials. Treatment duration, antibiotics used, and definition of response to treatment varied substantially. Pooled sensitivity (67%, 95% CI 42, 85) and specificity (73%, 95% CI 58, 85) of trial-of-antibiotics versus mycobacteriology were below internationally-defined minimum performance profiles for tuberculosis diagnostics with substantial heterogeneity (I296% for sensitivity, 99% for specificity)and low QUADAS-2 quality assessments. My trial (NCT03545373) aimed to strengthen this weak evidence-base. I randomised(1:1:1) Malawian adults attending primary care for illness ≥2 weeks including cough with no immediate indication for hospitalisation, no recent antibiotic or tuberculosis treatment or prevention, to: azithromycin (500mg daily, 3 days) amoxicillin (1g three times/day, 5 days); or standard-of-care (SOC) with no immediate antibiotic. Sputum at enrolment and day 8 was tested using microscopy, Xpert MTB/RIF, and culture. Primary outcomes were day 8 specificity (percentage with symptom improvement among mycobacteriology-negative), and day 29 clinical outcomes (composite: death, hospitalisation or missed tuberculosis diagnosis). The secondary outcome was day 29 risk of resistant Streptococcus pneumoniae identified by culture of nasopharyngeal swabs. Afterscreening5825 adults,1583 were randomised of whom 6.3% (100/1583) had positive baseline mycobacteriology. Compared to SOC (79.1%of 530), trial-of-antibiotics improved tuberculosis specificity: azithromycin (527patients) vs. SOC difference +8.6% (95% confidence interval 3.9%, 13.3%); amoxicillin (526 patients) vs. SOC difference +8.8% (4.0%, 13.6%), but with extremely low sensitivity (10.7% azithromycin, 23.3% amoxicillin). Day 29 composite clinical outcomeswere similar (SOC 1.1%, azithromycin 1.1%, amoxicillin 2.1%). Compared to SOC (5.3%), proportions with day 29 resistant Streptococcus pneumoniaewere higher in azithromycin +2.5% (-0.5, 5.5), but similar in amoxicillin +0.2% (-2.9, 2.5)arms. Using standard decision-analysis tools I compared antibiotic prescription and Xpert MTB/RIFrequirements of 3 algorithms combining Xpert MTB/RIF and trial-of-antibiotics against Xpert alone in 100,000 hypothetical patients with tuberculosis prevalence of 5%. Antibiotic prescriptions (trial-of-antibiotics plus out-of-protocol) needed to identify one tuberculosis patient varied from 22.7 to 220.4 in combination algorithms, with minimal benefit over Xpert alone (NNS =22.5Xpert tests, plus 4.0antibiotic prescriptions). Trial-of-antibiotics exposed patients to considerable misclassification risks. Conclusion: Harms of trial-of-antibiotics likely outweigh benefits, with the poor diagnostic performance ,lack of additional clinical impact, high cost, and likely impact on antimicrobial resistance arguing strongly against routine prescription. National tuberculosis and antimicrobial stewardship programs should limit outpatient prescription of empirical broad-spectrum antibiotics to patients with strong clinical or microbiological indications. Future research should focus on strengthening diagnostics for tuberculosis and other respiratory pathogens, evaluating antibiotic-sparing tuberculosis diagnostic algorithms in less clinically stable patients, and antimicrobial stewardship programs targeting tuberculosis-related prescribing.

Undiagnosed tuberculosis (TB) remains the most common cause of HIV-related mortality, including among people living with HIV (PLHIV) starting antiretroviral therapy (ART). This thesis explores opportunities for reducing PLHIV mortality in sub-Saharan Africa. Firstly, a systematic review of eight Xpert MTB/RIF (Xpert) impact trials found that lack of Xpert impact on mortality was mainly due to higher empiric TB treatment rates in microscopy versus Xpert arms. Secondly, the Botswana XPRES trial evaluated the effect of an intervention package comprising (1) support for intensified TB case finding (ICF), (2) strengthened tracing to support retention, and (3) Xpert replacing sputum-smear microscopy on early (6-month) ART mortality. Strengthened ICF and retention were associated with about 23% lower 6- month mortality. No mortality benefit of Xpert replacing microscopy was observed. Thirdly, to identify PLHIV at highest risk of early ART mortality, CD4-independent and - dependent scores were derived from XPRES data and externally validated. Sensitivity of CD4-independent score ≥4 in predicting mortality (86%) was twice that of WHO stage alone (48%). Both CD4-independent score ≥4 and CD4-dependent score ≥5 had similar sensitivity but higher specificity than WHO-recommended advanced HIV disease criteria (i.e., CD4 <200/μL or WHO stage III/IV). Finally, a TB risk score for PLHIV was derived from XPRES data and validated on three external datasets, with the aim of increasing (1) detection of asymptomatic TB, and (2) sensitivity to exclude TB prior to TB preventive therapy. In the external datasets, TB risk score ≥2 had higher sensitivity (87-97%) than the WHO four-symptom screening rule (80- 94%) but lower specificity (12-58% versus 16-70%). In conclusion, strengthening TB screening and retention in care could reduce early ART mortality in sub-Saharan Africa. In addition, early mortality and TB risk scores could help clinicians better detect and differentiate risk and should be further evaluated.

Despite their growing use, there is limited literature on the design and analysis of stepped-wedge trials (SWTs). The design is characterised by some or all clusters experiencing the control condition follow by the intervention condition. This enables within-cluster comparisons, but confounds the intervention effect with secular trends. In this thesis, I aim to use statistical methodology to improve the design efficiency and identify robust methods of analysis. I provide a new formulation of a design effect for an SWT. From this, I identify that it is more efficient for a trial to begin after the first clusters switch to the intervention, and end before the final clusters switch to the intervention. SWTs are commonly analysed using a mixed-effect model with a random effect for cluster and fixed effects for periods and the intervention. Through a wideranging simulation study, I found that this “standard” model is sensitive to deviations from model assumptions and suggest adding a random effect for period. However, this alternative model still suffered from confidence interval under coverage in some scenarios. I introduce a novel method of analysis that excludes the within-cluster comparisons. Each period of the trial is analysed separately to give within-period intervention effect estimates. An inverse-variance weighted average provides an overall effect and permutation tests provide a p-value and confidence intervals. In a simulation study, I find that this novel method provides unbiased inference in a range of scenarios, but had lower power than the standard model when the standard model was correctly specified. I introduce a new Stata command I have developed to conduct this novel method in order to encourage wider use of this new methodology. These finding will lead to trialists using more efficient trial designs and more robust analysis methods.

HIV-associated tuberculosis (HIV/TB) was responsible for an estimated 374,000 deaths globally in 2016. Much of this burden resides in hospitals in sub-Saharan Africa, where HIV/TB is usually disseminated and is the major cause of admission and death. Recently, urine-based detection of mycobacterial lipoarabinomannan (LAM) or nucleic acids (using the Xpert MTB/RIF assay) has improved diagnosis of HIV/TB in this population, with the potential to improve outcomes. However, disseminated 'urine-positive' TB may also be associated with a higher mortality risk and impaired immune responses compared to patients with HIV/TB disease who are urine TB test negative. This thesis addresses the hypothesis that positive urine-diagnostics in inpatients with HIV/TB disease are associated with mortality, can identify patients with impairment of immune responses, and can be used as useful prognostic markers for identifying patients with poor outcomes. First, a systematic review and meta-analysis was undertaken to synthesise existing evidence of the association between urine-LAM detection and mortality risk in HIV/TB (Gupta-Wright et al, BMC Med 2016). Ten eligible studies were identified, and random-effects meta-analysis demonstrated urine-LAM positive patients had a 2.3-fold greater mortality risk, and 2.5-fold greater adjusted odds of death than urine-LAM negative HIV-TB patients. Secondly, a prospective observational cohort study nested within a large randomised trial of urine-based TB screening was undertaken (STAMP trial in Malawi and South Africa, Gupta- Wright at al BMC Inf Dis 2016 and Lancet 2018). It included 322 patients with laboratoryconfirmed HIV/TB disease and demonstrated a remarkably high (31%) 2-month mortality risk despite rapid initiation of TB treatment. It also demonstrated that advanced immunosuppression was common in HIV/TB disease despite high antiretroviral therapy (ART) coverage. Cohort data were used to identify clinical phenotypes associated with poor outcomes: urine-test positivity (LAM, Xpert or both) was independently associated with a 50% increase in 2 month case-fatality. Thirdly, a study of immune responses was nested in the Malawi STAMP trial site. A functional whole blood assay of phagocytic activity was developed (Gupta-Wright et al, Frontiers Immunology 2017) and utilised in 65 HIV/TB patients and 16 matched HIV-positive TB-negative controls. Cellular and soluble markers of immune activation, ex-vivo monocyte and T-cell cytokine responses and multiplex plasma cytokine and chemokine levels were also measured. Poor outcomes and urine-positive HIV/TB disease were associated with broadly impaired immune responses, including phagocytic oxidative burst function, monocyte activation and dominant innate immune responses. Finally, urine-LAM was included in a simple, pragmatic clinical score to identify patients at high risk of a poor outcome in resource-poor settings, which was externally validated on an individual-patient record dataset combining data from 2 different studies over 5 African countries (Gupta-Wright et al, PLOS Medicine 2019). This PhD has demonstrated the high prevalence and case-fatality associated with disseminated HIV/TB disease in HIV-positive hospital admissions despite widespread access to ART, and the utility of urine-diagnostics in identifying patients at high risk of mortality, in addition to their diagnostic use. These linked studies could potentially inform on the choice and nature of interventions to reduce the high mortality of HIV/TB, including better identification and management of ART failure, therapies to address TB-related immune dysfunction, and improved supportive care and prevention and treatment of co-morbidities.

Improved availability of HIV tests has led to increases in numbers testing and starting treatment in sub-Saharan Africa. Despite such remarkable progress, men continue to lag behind in HIV testing in the region including men in well-established heterosexual relationships, in which context HIV transmission is surprisingly high. We previously found HIV self-testing (HIVST) to be very effective at increasing the uptake of HIV testing in the general population in urban Blantyre, Malawi. This PhD investigated the effect of partner-delivered HIVST, providing HIVST kits to pregnant women in antenatal clinics (ANC) with or without additional interventions, including financial incentives, on uptake of testing and linkage to care or prevention. The PhD is made up of three main pieces of work: First, a systematic review was conducted to investigate the existing evidence regarding the effectiveness of demand-side (given to users) financial incentives on linkage to HIV treatment or voluntary medical male circumcision (VMMC) in low and middle income (LMIC) countries. Relevant electronic databases and conference proceedings were searched for randomised controlled trials. Seven trials were identified out of 1099 citations, with all showing significant improvement in linkage: four investigated VMMC and three investigated ART. Manuscript currently under review. Secondly, a formative study was carried out to identify additional potential interventions and to refine interventions identified as promising through the systematic review, before being tested in a subsequent trial. Undertaking this formative study ensured that interventions being considered for inclusion in the trial design were adapted to the local environment and prevailing social norms, by seeking input and feedback from would-be users of the service. Paper published in J Int AIDS Soc, 2017. Thirdly, a multi-arm two-stage cluster-randomised trial was conducted in Blantyre, Malawi. The paper describing the trial design is published in Trials, 2017; trial results manuscript is under review. Antenatal care clinic days were randomized to standard of care (SOC: personalised invitation to male friendly clinic for standard HIV testing and fast-track referral for HIV treatment or VMMC services) or one of five intervention arms: SOC plus two partner-delivered self-test kits with a) no addition, or financial incentives of b) US$3, c) US$10, d) lottery (10% chance of winning $30), or e) phone call. All incentives were conditional on attending the male friendly clinic. The primary outcome at 28 days, measured through attendance at the male friendly clinic, was: referral for antiretroviral therapy (ART) for HIV-positive men; or voluntary male medical circumcision (VMMC) scheduled if HIV-negative/uncircumcised; or counselling if HIV-negative/circumcised. At the end of stage 1, a planned interim analysis was performed and the HIVST-lottery arm was dropped for futility. Male partner HIV-testing was substantially increased in all HIVST arms (range 87.0% to 95.4% in the 5 arms, compared to 17.4% in the SOC arm), according to self-report by the woman at 28 days. Reaching the primary linkage outcome at 28 days was most likely for the partners of participants in clinic days randomised to the HIVST-$3 and the HIVST-$10 arms, with geometric means of 40.9% (adjusted risk ratio [aRR] 3.01, 95%CI:1.63-5.57) and 51.7% (aRR 3.72, 95%CI:1.85-7.48), respectively. Successful male linkage was also more likely in the HIVST-phone reminder (geometric mean 22.3%, aRR 1.58, 95%CI:1.07-2.33) and HIVST-alone (geometric mean 17.5%: aRR 1.45 (95%CI:0.99-2.13) compared to SOC (13.0%). Linkage in the HIVST-lottery arm (geometric mean 18.6%, aRR 1.43, 95%CI:0.96-2.13) was less pronounced than with the $3 or $10 fixed conditional-incentives, and clients disliked the uncertainty. Overall, 42/46 (91.3%) newly diagnosed HIV-positive men initiated ART and 135/222 (60.8%) HIV-negative and previously uncircumcised men had VMMC. No serious adverse events were reported. Cost per male partner attended clinic with confirmed HIV test result was $23.73 and $28.08 for $10 and $3 arms, respectively. Secondary distribution of HIVST kits from ANC clinics greatly increased partner-testing, and timely linkage within 28 days increased 3-fold with the combination of fixed financial incentives plus partner-delivered HIV self-test kits in this hard to reach group. This PhD project has demonstrated that novel trial designs such as adaptive MAMS can be applied to address pressing public health problems in Africa. The approach followed here, combining systematic review, qualitative pilot study, and multi-arm randomised trial is ideal for rapidly generating high quality evidence for interventions, such as financial incentives, where the effectiveness of different amounts may vary from one setting to the next.

Phase III trials for new tuberculosis treatment regimens require large numbers of participants and can take over five years to complete. A surrogate marker for poor outcome (failure at end of treatment or recurrence following successful treatment), the established endpoint in such trials, could shorten trial duration and reduce trial size. Culture results after two months of treatment have shown the most promise but, prior to this research, no formal evaluation had been performed. In this thesis, culture results during treatment are evaluated as prognostic and surrogate markers for poor outcome using data on 6974 patients from twelve tuberculosis treatment randomised controlled multi-arm trials conducted in East Africa and East Asia. A strong association was found between culture results during treatment and poor outcome. Nevertheless, culture results were not good patient-specific predictors of poor outcome with low sensitivities and specificities. Existing meta-analytic methods for evaluating surrogate markers are not wholly suited to this setting of multi-arm trials with binary true and surrogate endpoints. Extending these methods, the two month culture was found to be a good surrogate marker using data from Hong Kong trials and the three month culture was found to be a good surrogate marker using data from East African trials. These results are an indication that cultures during treatment do capture some of the treatment effect. Further work is needed in understanding the differences between the Hong Kong and East African trials. The meta-analytic methods for evaluating surrogate markers in this thesis included a graphical representation that permitted a clear visual evaluation of the surrogate. Methods developed in this thesis for modelling the relationship between the treatment effects on the true and surrogate endpoints were not satisfactory. The deficiencies were not overcome with the two extensions proposed. Further work is needed in developing a more appropriate model.