The effects of longitudinal HIV viral load exposure on immune outcomes, mortality, and opportunistic infections in people on ART in sub-Saharan Africa

Sempa, Joseph Bukulu (2017-12)

Thesis (PhD)--Stellenbosch University, 2017.

Thesis

ENGLISH SUMMARY : Introduction: Longitudinal viral load monitoring is used as a cross-sectional marker for treatment failure in HIV infected people receiving antiretroviral therapy. Cumulative viral load, as quantified by area under the viral load curve during combination antiretroviral therapy, has been correlated with treatment outcomes in studies outside, but not within, sub-Saharan Africa. We investigate the effects of exposure to longitudinal viral load on, the incidence of opportunistic infections, mortality and immune recovery in local, previously combination antiretroviral therapy naïve, cohorts. Further, we systematically review statistically derived immune response models and use this to define priors for Bayesian models for application on a previously undescribed treatment cohort. Methods: We analyze data from the Infectious Diseases Institute (IDI) cohort, Kampala-Uganda, and the Antiretroviral Clinic at Tshwane District Hospital in Gauteng-South Africa. For the systematic review, we use ‘Preferred Reporting Items for Systematic Review and Meta- Analyses’ guidelines. We also compare cumulative viral load as numerically estimated using two methods: area under the viral load curve, which is then log-transformed, named, ‘untransformed cumulative viral load’; and area under the log-transformed viral load curve, above the kit-based detection limit of 400 copies/mL, named, ‘transformed cumulative viral load’. We use Cox Proportional Hazards and Bayesian Generalized Mixed Effects to define treatment outcome models. Results: In the IDI cohort most recent viral load, not cumulative viral load, is associated with a 1.34-fold (95% confidence interval: 1.12, 1.61) increase in the risk of opportunistic infections. Transformed, not untransformed, cumulative viral load is associated with mortality and immune response. Each log10 copy-yr/mL increase corresponds to a 1.63-fold (95% confidence interval: 1.02, 2.60) increase in risk of mortality. Systematic review of immune response statistical models also reveals many differences in the number and type of variables adjusted-for, variable transformations and scales and scant details regarding the modelling methods employed. In the Tshwane cohort, using Bayesian methods, for the slope of longitudinal CD4 counts, each log10 copy-yr/mL increase cumulative viral load corresponds to a mean annual CD4 count decrease of -19.5 cells/μL (95% credible interval: -28.34, -10.72). Further, in the asymptote model, each log10 copy-yr/mL increase reduced the odds of having a CD4 count ≥500 cells/μL to 0.42 (95% credible interval: 0.242, 0.724). Modelling inherently variable absolute CD4 count using a Student’s t-distribution produced better fits than assuming a Gaussian normal distribution. Discussion: Transformed cumulative viral load is associated with both mortality and long-term immune response, while most recent viral load is associated with incidence of opportunistic infections. This thesis emphasizes the need for the review of existing literature prior to any statistical analyses, so that more comparable and robust statistical models than have been available to date will be constructed. In particular, comparing immunological outcomes (CD4 counts), statistical models for sub-Saharan African cohorts would benefit from the application of more uniform modelling techniques. Adjusting for transformed cumulative viral load and the use of appropriate distributional assumptions, improves the modelling of immune response to antiretroviral therapy. Future statistical immune response models would benefit from the use of Bayesian methods owing to their flexibility in the selection of prior distributions and hierarchical model designs.

AFRIKAANSE OPSOMMING : Inleiding: Die monitering van longitudinale viruslading word gebruik as 'n biomerker vir behandelingsfaling in HIV-geïnfekteerde mense wat kombinasie antiretrovirale terapie ontvang. Kumulatiewe viruslading, gekwantifiseer as die area onder die viruslading kurwe tydens terapie, is gekorreleer met behandelingsuitkomste in studies elders, maar nie in Afrika suid van die Sahara. In hierdie studie word die effek van longitudinale viruslading, insluitend die voorkoms van opportunistiese infeksies, sterfte en die herstel van die immuunstelsel in plaaslik behandelde pasiënte ondersoek. Verder, word ʼn sistematiese oorsig van statistiese immuunrespons modelle uitgevoer. Hierdie resultate word gebruik om Bayesiaanse modelle te definieer, vir toepassing op 'n voorheen onbeskrewe pasiënt groep. Metodes: Kohorte van die Infectious Diseases Institute (IDI), in Kampala, Uganda, en van die Antiretrovirale Kliniek by die Tshwane Distrikshospitaal in Gauteng, Suid-Afrika is geanaliseer. Vir die sistematiese oorsig gebruik ons die sogenaamde ‘Preferred Reporting Items for Systematic Review and Meta-Analyses’ riglyne. Ons vergelyk ook kumulatiewe virusladings beraam met twee numeriese metodes: 1) die log-getransformeerde area onder die viruslading kurwe, genaamd 'ongetransformeerde kumulatiewe viruslading'; en 2) die area onder die log-getransformeerde viruslading kurwe, bo die toets-spesifieke limiet van deteksie van 400 kopieë/ml, genaamd die 'getransformeerde kumulatiewe viruslading'. Ons gebruik ‘Cox Proportional Hazards’ en ‘Bayesian Generalized Mixed Effects’ om behandelingsuitkoms modelle te definieer. Resultate: Vir die IDI kohort is die mees onlangse viruslading, i.e. die nie-kumulatiewe viruslading, geassosieer met 'n 1.34-voud toename (95% vertrouensinterval: 1.12, 1.61) in die risiko van opportunistiese infeksies. Die getransformeerde kumulatiewe viruslading is geassosieer met sterfte en immuunrespons. Elke log10 kopie/jr/ml verhoging stem ooreen met 'n 1.63-voud toename (95% vertrouensinterval: 1.02, 2.60) in die risiko vir sterfte. Die sistematiese oorsig van statistiese modelle vir immuunrespons het ook baie verskille getoon in die aantal en tipe veranderlikes waarvoor aangepas was, veranderlike transformasies en skale, en besonderhede was skaars oor die modelleringsmetodes wat gebruik was. In die Tshwane kohort, beraam met behulp van Bayesiaanse metodes, veroorsaak elke log10 kopie/jr/ml kumulatiewe viruslading toename 'n gemiddelde jaarlikse CD4-telling afname van -19.5 selle/μL (95% vertrouensinterval: -28.34, -10.72). Verder, in die asimptootmodel, verminder elke log10 kopie/jr/ml toename in viralelading die kans om 'n CD4-telling van meer as 500 selle/μL met 0.42 (95% vertrouensinterval: 0.242, 0.724). Die modellering van inherente veranderende CD4 telling het met behulp van 'n Student se t-verdeling beter modelle geproduseer as vir Gaussiese normaal verdelings. Bespreking: Getransformeerde kumulatiewe viruslading is geassosieer met beide sterfte en langtermyn immuunrespons, terwyl die mees onlangse viruslading verband hou met die voorkoms van opportunistiese infeksies. Hierdie proefskrif beklemtoon die vereiste vir ʼn oorskou van bestaande literatuur voordat enige statistiese ontledings onderneem word, sodat meer vergelykbare en robuuste statistiese modelle gebou sal word as wat tot dusver beskikbaar was. Die vergelyking van immuunrespons (CD4-telling) statistiese modelle in antiretrovirale terapie sal baat vind by die toepassing van meer eenvormige modelleringstegnieke. Sulke modelle is verbeter deur gebruik van getransformeerde kumulatiewe viruslading en meer akkurate verdelingsaannames. Toekomstige statistiese immuunrespons modelle sal ook baat vind by die gebruik van Bayesiaanse metodes as gevolg van hul aanpasbaarheid in terme van verdelings keuses en die implementasie van hiërargiese modelontwerpe.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/102575
This item appears in the following collections: