Factors associated with World Health Organisation (WHO) clinical staging and related characteristics in HIV positive patients: an application of multistate, missing data and modelling techniques

Simple item page
dc.contributor.authorBatsirai, Murapa
dc.date.accessioned2019-09-12T12:32:21Z
dc.date.available2019-09-12T12:32:21Z
dc.date.issued2019
dc.description.abstractBackground Human Immunodeficiency Virus (HIV) remains a significant problem in sub-Saharan Africa which has the highest number (25.6 million) of people living with HIV (PLHIV). South Africa is amongst the top rank of sub-Saharan Africa countries with the highest HIV prevalence. Many studies have been done to have an in-depth understanding of HIVandAIDSdiseaseresultinginvariousinterventionsbeingimplementedtoimprove the lives of people infected by the disease. These studies are usually done using longitudinaldesignswhichhavetheadvantageofenablingresearcherstoobservepatient changes (outcomes) over time; however, they are prone to missingness due to unobserved data as patients may miss scheduled visits. This study aims to determine transition probabilities between WHO stages I, II, III and IV over time and compare Rubin’s and Bayesian methods in determine factors associated with WHO stage ailments and symptomatic conditions amongst HIV infected patients on patient level data from the Adult Wellness study. Methods The researcher conducted a secondary data analysis of the Adult Wellness study which was conducted from 2002 to 2010, to be able to quantify changes in ailments and symptomatic conditions over time, the researcher fitted the general multi-state Markov model which assumes that patients may develop more severe ailments and symptomaticconditions. ThestatesweredefinedbasedonWHOstages,thatis,stage I, II ,III and IV. The researcher also fitted three random effects ordered logistic regression models to determine factors associated with these WHO stage outcomes. The researcher employed the maximum likelihood estimation (MLE) on the first model fittedonrawdataandsecondmodelaftermultipleimputation(MI)toaccountformissing data. The last model adopted Bayesian estimation (BE) to the raw data. Finally, the researcherperformedasensitivityanalysisusingsimulateddataandfittedallthethree models described earlier. Results A total of 2,609 patients accounted for 12,102 observations were analysed. Majority of the patients were females (77.4%) antiretroviral therapy ART naïve (61.5%) having attained Grade 0-12 (77.9%). The Markov multi-state model showed that patients in WHO stage II were 1.33 times more likely to move to WHO stage III than WHO stage I whilst patients in WHO stage III were 2.26 times (0.16118/0.07124) more likely to move to WHO stage II than progressing to WHO stage IV. The probability of remaining in WHO stage I was 59% after eight year follow up period. Relative to patients with no ailments and symptomatic conditions, patients with one HIV ailments or symptomatic condition has an INCREASED RISK of progressing to advanced WHO stages, (model with raw data OR 2.07, 95%CI: 1.23-3.30). There were some unexpected results were patients on ART and cotrimoxazole (CTX) drugs showed to have increased odds of becoming worse than their counterparts. This was evident in all the three models: raw data MLE model OR 1.5828 (95% CI 1.1948,2.0969) and OR 2.0670 (1.5619,2.7355); MI MLE model OR 1.2252 (95% CI 1.0884,1.3793) and OR 1.5438 (95% CI 1.3186,1.8074); and raw data BE model OR 1.6096 (95% CI 1.2055,2.0952) and OR 2.0758 (1.5507,2.7270) for ART and CTX respectively. Both MLE and BE for the raw data gave similar estimates; however, these estimates were different from the MI MLE model which were more precise (smaller standard errors). Conclusions The results showed that patients had an increased chance of remaining in the same state than either advancing in WHO stages of ailments and symptomatic conditions or recovering. If the level of missing data is reasonable, it is recommended to apply the MI techniques. Multiple imputations and Bayesian missing data methods should be used together and determine which one produce better results per each situation. Finally simulated results showed that multiple imputation and Bayesian models become different as the percentage of missing data increasesen_ZA
dc.description.librarianMT 2019en_ZA
dc.format.extentOnline resource (91 leaves)
dc.identifier.citationMurapa, Batsirai Amen (2019) Factors associated with world health organisation (WHO) clinical staging and related characteristics in HIV positive patients:an application of multistate, missing data and modelling techniques, University of the Witwatersrand, Johannesburg, <http://hdl.handle.net/10539/28094>
dc.identifier.urihttps://hdl.handle.net/10539/28094
dc.language.isoenen_ZA
dc.subject.meshHIV infections
dc.subject.meshAIDS (Disease)
dc.subject.meshHIV infections--Social conditions
dc.titleFactors associated with World Health Organisation (WHO) clinical staging and related characteristics in HIV positive patients: an application of multistate, missing data and modelling techniquesen_ZA
dc.typeThesisen_ZA
Files
Original bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
1734164-Batsirai Murapa.pdf
Size:
1.06 MB
Format:
Adobe Portable Document Format
Description:
Download
License bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
1.71 KB
Format:
Item-specific license agreed upon to submission
Description:
Download
Collections
ETD Collection