Introduction
HIV is a viral infection that severely impairs the immune system.1 Since the introduction of antiretroviral therapy (ART) in the mid-1990s, it has led to significant reductions in HIV-associated morbidity and mortality, transforming HIV infection into a manageable chronic condition.2–6
Despite these advancements, HIV/AIDS continues to impose significant disease burdens worldwide, with 38.4 million people living with HIV (PLHIV) and 650 000 AIDS-related deaths in 2021.7 The increasing life expectancy of PLHIV presents new challenges in the treatment and management of this expanding population. Studies indicate that PLHIV faces a higher risk of developing concurrent chronic diseases, including cardiovascular disease and non-AIDS-related neoplasms, compared with the general population.8–10 This heightened risk may be related to HIV-induced immunodeficiency, ART drug toxicities, ageing and poor lifestyle factors.8 9 Moreover, while the introduction of regimens such as integrase strand transfer inhibitor (INSTI) in recent decades has undoubtedly represented a significant advancement in treatment of HIV/AIDS, it is also crucial to acknowledge that these drugs have presented certain clinical complexities. A recent multicenter study revealed that the initiation of INSTIs was associated with an early-onset, excess incidence of cardiovascular disease in the first 2 years of exposure, after accounting for known cardiovascular disease risk factors.11 The proportion of deaths from non-AIDS-related illnesses among PLHIV has surpassed that of AIDS-related illnesses in many regions of the world where free ART is available.12 13 Furthermore, recent real-world studies have raised doubts about the efficacy of new drugs such as INSTIs, as no significant disparities in mortality rates of PLHIV were observed among different modern ART regimens.14–16 In light of this scenario, governments and health authorities worldwide must continue to allocate significant medical resources to the treatment and management of HIV-infected populations each year.
Developing an ideal prognostic model for treatment experienced people living with HIV (TE-PLHIV) is crucial to optimising HIV care and providing tailored treatment for individual patients, ultimately improving treatment outcomes and facilitating the rational allocation of limited health resources.17 Given the identification of multiple risk factors associated with HIV/AIDS-related mortality, combining several independent indicators could significantly enhance predictive effectiveness. Existing studies of the prognostic model for TE-PLHIV have identified several demographic and laboratory indicators closely linked to the mortality of TE-PLHIV, including age, sex, WHO clinical stage, body weight or body mass index (BMI), CD4+ T cell count, HIV viral load and haemoglobin.18–30 However, there are discrepancies in the characteristics of prognostic models of TE-PLHIV, resulting in variations in the selected demographic and clinical features, as well as in the parameters and performance. Publication bias, poor reporting and extensive heterogeneity are recognised issues in prognostic research.31–33 Despite the number of prognostic prediction models established for TE-PLHIV, none of them have been widely accepted as an addition to precision medicine workflows. A systematic review and meta-analysis of these models are therefore essential for enhancing their methodological quality, including better indicators and performance.
Study aim
The aim of this systematic review is to identify, screen and evaluate all published prognostic prediction models on survival outcomes of TE-PLHIV and inform clinical therapeutic decision-making.
Methods
Study design
The systematic review described in this article is scheduled to be conducted between January 2024 and January 2025. This review was registered on the PROSPERO international registry of systematic reviews on 8 April 2023 (CRD42023412118).
This protocol adheres to a number of reporting guidelines for a systematic review including the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocols (PRISMA-P) guideline34 and Cochrane Prognosis Methods Group Protocol Template35 and guidelines for assessing and reporting prediction model studies including transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) statement,36 PROBAST tool (prediction model risk of bias assessment)37 and the checklist for corresponding CHARMS checklist (CHecklist for critical Appraisal and data extraction for systematic Reviews of prediction Modelling Studies).38 Key items of this review are clarified with assistance of the PROGnosis RESearch Strategy (PROGRESS) Partnership (table 1).
Framing of this systematic review with key items identified by the CHARMS checklist
Eligibility criteria
Any study design including primary research (eg, randomised controlled trial, cohort study and case–control study) that reports on one or more statistical models, tools or scores with at least two predictors proposed to predict a TE-PLHIV’s risk of a future survival outcome (prognostic prediction modelling studies) before January 2024 will be considered for inclusion in the review. Prognostic prediction modelling studies can be either model development, model validation or a combination. Studies that have a single risk factor in the model will be excluded, as they are limited in their utility for individual risk prediction. Editorial comments or letters will be excluded from the review. Eligible studies included in the review will be limited to those conducted in humans by applying the Cochrane Group’s filter for Humans not Animals filter.39 PICOTS (population, index, comparator, outcome, timing, setting) approach will be used to frame the eligibility criteria and to guide the selection of prognostic prediction modes (table 2).33 37
Eligibility criteria for the systematic review framed with the PICOTS system
Databases
A comprehensive and systematic literature search will be conducted in five major publicly available electronic medical literature databases: (1) PubMed (covering the period from 1946 to present); (2) EMBASE (covering the period from 1947 to present); (3) Scopus (covering the period from 1996 to present); (4) the Cochrane Library; and (5) OpenGrey (www.opengrey.eu/), which is a database containing conference abstracts that have not been published.
Search strategy
The search strategy was built using keywords including HIV-related terms (ie, ‘HIV’, ‘human immunodeficiency virus’ and ‘HIV infection’), AIDS-related terms (ie, ‘AIDS’ and ‘acquired immunodeficiency syndrome’), patient-related terms (ie, ‘people living with HIV/AIDS’ and ‘PLHIV’) and prediction modelling-related terms (ie, ‘prognostic tool’, ‘prognostic model’, ‘prognostic nomogram’, ‘prognostic value’, ‘prognostic risk’, ‘prognostic score’, ‘prognostic scoring’, ‘prognostic marker’, ‘disease progression score’, ‘risk index for mortality’ and ‘multivariate models for predicting progression’). The draft search strategy is provided in the online supplemental appendix. The search strategy, specifically, subject indexing terms will be translated appropriately for the Cochrane Library and OpenGrey. The final search strategy will be iteratively refined.
Supplemental material
The reference lists of included model development studies and relevant systematic reviews for further studies will be hand searched for additional potentially relevant citations. We will also search for publications which cited the model development studies to identify model validation studies. The included studies will be checked for error or fraud. We will not place any restrictions on language, publication year or publication status when searching the electronic databases. Any non-English studies identified will be translated and assessed for eligibility.
Data collection and analysis
Data management
Retrieved studies included in the review will be imported into Endnote reference manager software (V. 20.4.1, Clarivate Analytics, Philadelphia, USA, available at https://endnote.com/). Duplicate records will be identified and excluded using a systematic, rigorous and reproducible method using a sequential combination of fields including author, year, title, journal and pages.40 Screened records will then be imported into Covidence systematic review software throughout the review (Veritas Health Innovation, Melbourne, Australia, available at http://www.covidence.org).
Selection process
Two reviewers (XW and YC) will independently screen and review the abstracts of all studies identified by the final search strategy. For each excluded article, reasons for exclusion will be specified. Disagreement or doubt will be resolved by consensus, and if consensus cannot be reached, the full texts of the study will be independently accessed for further evaluation. Any conflict will be resolved through discussion with a senior advisor (HZ), if necessary.
Data extraction
Two reviewers (XW and YC) will independently extract data from eligible studies included in the review, using a standardised electronic form developed with reference to the checklist for CHARMS.38
For each eligible study, we plan to seek information on objective, source of data, participants, survival outcome(s) to be predicted, candidate predictors, sample size, missing data, model development, model performance (discrimination, calibration, clinical utility and measures of case-mix variation), results including final multivariable models and interpretation of presented models and model validation.38 Information not reported in the publications will be obtained from the authors wherever possible, in addition, if insufficient information is obtained the study will be excluded from the review. Any disagreement will be resolved through consultation with a senior advisor (HZ).
Assessment of risk of bias
The methodological quality (risk of bias) and relevance (applicability) reporting the development or validation of a prognostic model will be systematically assessed using the PROBAST.37 This tool is structured around four key domains: participant selection (eg, were inclusions and exclusions of participants appropriate); predictors (eg, were predictors measured blind to outcome data); outcomes (eg, was the same definition used for outcomes in all patients); sample size and patient flow (eg, was there a prespecified sample size based on the estimated number of outcome events and handling of missing data); and analysis (eg, was selection of predictors based on univariable analysis avoided). Each study will be given a rating of high, low or unclear risk for each of the four domains.
Synthesis
All included studies on prognostic prediction models will be described, with key findings tabulated to facilitate comparison of survival outcomes to be predicted, predictors included in the final model and performance measures.38 Measures of uncertainty will be reported when published or approximated using published methods.33 Where reported, classification measures such as sensitivity, specificity, positive predictive value and negative predictive value will also be included.38
When at least five studies report similar measures of association, a meta-analysis will be conducted for each prognostic factor with reference to the Meta-analysis of Observational Studies in Epidemiology group guidelines.41 The restricted maximum likelihood and Hartung–Knapp–Sidik–Jonkman methods will be used to estimate the between-study heterogeneity and 95% CI for the average performance.33 Statistical or clinical homogeneity will be assessed using the I2 statistics where I2 values >50% indicate moderate to high heterogeneity.42 When it is not feasible to conduct a quantitative synthesis, the evidence will be summarised narratively.
Subgroup analyses
Where there are enough eligible studies included in the review, we will conduct subgroup analyses of the type of prognostic prediction modelling studies (ie, development or validation), target population to whom the model applies, whether the population was treated (yes/no), treatment type after ART initiation, the follow-up duration, survival outcomes to be predicted and study quality (risk of bias).
Sensitivity analysis
A sensitivity analysis is a repeat of a meta-analysis. To test the robustness of this review’s findings, we would undertake sensitivity analyses to explore the influence on model performance for exclusion of studies at lower and higher risk of bias.33
Reporting and presentation of findings
Reporting and presentation of findings will be guided by the PRISMA statement34 and relevant recommendations from the TRIPOD statement.36 The grading of recommendations, assessment, development and evaluation approach will be used to determine confidence in estimates.43–45 The assessments will be performed by two independent reviewers (XW and YC) independently. Disagreements will be resolved by discussion or third-party adjudication (HZ), when necessary.
Amendments
Protocol amendments will be listed and made available on the PROSPERO registration. Date, description and rationale will be given for each amendment.
This post was originally published on https://bmjopen.bmj.com