Personalised risk prediction tools for cryptococcal meningitis mortality to guide treatment stratification; a pooled analysis of two randomised-controlled trials

Personalised risk prediction tools for cryptococcal meningitis mortality to guide treatment stratification; a pooled analysis of two randomised-controlled trials | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Personalised risk prediction tools for cryptococcal meningitis mortality to guide treatment stratification; a pooled analysis of two randomised-controlled trials View ORCID Profile T.H.A. Samuels , View ORCID Profile S.F. Molloy , View ORCID Profile D.S. Lawrence , View ORCID Profile A. Loyse , C. Kanyama , View ORCID Profile R.S. Heyderman , W.S. Lai , View ORCID Profile S. Mfinanga , S. Lesikari , View ORCID Profile D. Chanda , View ORCID Profile C. Kouanfack , View ORCID Profile E. Temfack , View ORCID Profile O. Lortholary , View ORCID Profile M.C. Hosseinipour , View ORCID Profile A.K. Chan , View ORCID Profile D.B. Meya , View ORCID Profile D.R. Boulware , View ORCID Profile H.C. Mwandumba , View ORCID Profile G. Meintjes , View ORCID Profile C. Muzoora , M. Mosepele , View ORCID Profile C.E. Ndhlovu , View ORCID Profile N. Youssouf , View ORCID Profile T.S. Harrison , View ORCID Profile J.N. Jarvis , View ORCID Profile R.K. Gupta doi: https://doi.org/10.1101/2024.07.10.24310212 T.H.A. Samuels 1 Division of Infection and Immunity, University College London - London ( United Kingdom ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for T.H.A. Samuels S.F. Molloy 2 Centre for Global Health, Institute for Infection and Immunity - St. George’s University of London - London ( United Kingdom ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for S.F. Molloy D.S. Lawrence 3 Department of Clinical Research, Faculty of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine - London ( United Kingdom ) 4 Botswana Harvard Health Partnership - Gaborone ( Botswana ) 5 School of Pathology, Faculty of Health Sciences, University of the Witwatersrand – Johannesburg ( South Africa ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for D.S. Lawrence A. Loyse 2 Centre for Global Health, Institute for Infection and Immunity - St. George’s University of London - London ( United Kingdom ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for A. Loyse C. Kanyama 6 The University of North Carolina Project – Lilongwe ( Malawi ) 7 Kamuzu Central Hospital , Lilongwe - Lilongwe ( Malawi ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site R.S. Heyderman 1 Division of Infection and Immunity, University College London - London ( United Kingdom ) 8 Malawi–Liverpool–Wellcome Trust Clinical Research Programme - Blantyre ( Malawi ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for R.S. Heyderman W.S. Lai 9 NIHR Biomedical Research Centre Clinical and Research Informatics Unit, University College London Hospitals NHS Foundation Trust – London ( United Kingdom ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site S. Mfinanga 10 The National Institute for Medical Research, Muhimbili Medical Research Centre - Dar Es Salaam ( Tanzania ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for S. Mfinanga S. Lesikari 10 The National Institute for Medical Research, Muhimbili Medical Research Centre - Dar Es Salaam ( Tanzania ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site D. Chanda 11 Institute for Medical Research and Training - Lusaka ( Zambia ) 12 University Teaching Hospital - Lusaka ( Zambia ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for D. Chanda C. Kouanfack 13 University of Dschang - Dschang ( Cameroon ) 14 Hôpital Central Yaoundé/ Site Agence Nationale de Recherche sur le Sida (ANRS) Cameroun , - Yaoundé ( Cameroon ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for C. Kouanfack E. Temfack 15 Molecular Mycology Unit, Institut Pasteur - Paris ( France ) 16 Douala General Hospital - Douala ( Cameroon ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for E. Temfack O. Lortholary 17 Paris Descartes University - Paris ( France ) 18 Necker Pasteur Center for Infectious Diseases and Tropical Medicine - Paris ( France ) 19 Assistance Publique–Hôpitaux de Paris - Paris ( France ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for O. Lortholary M.C. Hosseinipour 6 The University of North Carolina Project – Lilongwe ( Malawi ) 7 Kamuzu Central Hospital , Lilongwe - Lilongwe ( Malawi ) 20 University of North Carolina, Chapel Hill - North Carolina ( United States ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for M.C. Hosseinipour A.K. Chan 21 The Division of Infectious Diseases, Department of Medicine, Sunnybrook Health Sciences Centre, University of Toronto - Toronto ( Canada ) 22 Dignitas International, Zomba Central Hospital - Zomba ( Malawi ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for A.K. Chan D.B. Meya 23 Infectious Diseases Institute, College of Health Sciences, Makerere University - Kampala ( Uganda ) 24 University of Minnesota - Minneapolis, Minnesota ( United States ) 25 Department of Medicine, School of Medicine, Makerere University - Kampala ( Uganda ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for D.B. Meya D.R. Boulware 24 University of Minnesota - Minneapolis, Minnesota ( United States ) 25 Department of Medicine, School of Medicine, Makerere University - Kampala ( Uganda ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for D.R. Boulware H.C. Mwandumba 8 Malawi–Liverpool–Wellcome Trust Clinical Research Programme - Blantyre ( Malawi ) 26 Department of Medicine, Kamuzu University of Health Sciences - Blantyre ( Malawi ) 27 Liverpool School of Tropical Medicine - Liverpool ( United Kingdom ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for H.C. Mwandumba G. Meintjes 28 Wellcome Centre for Infectious Diseases Research in Africa (CIDRI-Africa), Institute of Infectious Disease and Molecular Medicine, University of Cape Town - Cape Town ( South Africa ) 29 Department of Medicine, University of Cape Town - Cape Town ( South Africa ) 30 Blizard Institute, Faculty of Medicine and Dentistry, Queen Mary University of London - London ( United Kingdom ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for G. Meintjes C. Muzoora 31 Mbarara University of Science and Technology - Mbarara ( Uganda ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for C. Muzoora M. Mosepele 4 Botswana Harvard Health Partnership - Gaborone ( Botswana ) 32 Department of Internal Medicine, University of Botswana - Gaborone ( Botswana ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site C.E. Ndhlovu 33 Internal Medicine Unit, Faculty of Medicine and Health Sciences, University of Zimbabwe - Harare ( Zimbabwe ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for C.E. Ndhlovu N. Youssouf 3 Department of Clinical Research, Faculty of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine - London ( United Kingdom ) 4 Botswana Harvard Health Partnership - Gaborone ( Botswana ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for N. Youssouf T.S. Harrison 2 Centre for Global Health, Institute for Infection and Immunity - St. George’s University of London - London ( United Kingdom ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for T.S. Harrison J.N. Jarvis 3 Department of Clinical Research, Faculty of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine - London ( United Kingdom ) 4 Botswana Harvard Health Partnership - Gaborone ( Botswana ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for J.N. Jarvis R.K. Gupta 34 UCL Respiratory, Division of Medicine, University College London – London ( United Kingdom ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for R.K. Gupta For correspondence: r.gupta{at}ucl.ac.uk Abstract Full Text Info/History Metrics Supplementary material Data/Code Preview PDF ABSTRACT Background Cryptococcal meningitis is a leading cause of adult community-acquired meningitis in sub-Saharan Africa with high mortality rates in the first 10 weeks post diagnosis. Practical tools to stratify mortality risk may help to tailor effective treatment strategies. Methods We pooled individual-level data from two randomised-controlled trials of HIV-associated cryptococcal meningitis across eight sub-Saharan African countries (ACTA, ISRCTN45035509 ; Ambition-cm, ISRCTN72509687 ). We used this pooled dataset to develop and validate multivariable logistic regression models for 2-week and 10-week mortality. Candidate predictor variables were specified a priori . ‘Basic’ models were developed using only predictors available in resource-limited settings; ‘Research’ models were developed from all available predictors. We used internal-external cross-validation to evaluate performance across countries within the development cohort, before validation of discrimination, calibration and net benefit in held-out data from Malawi (Ambition-cm trial). We also evaluated whether treatment effects in the trials were heterogenous by predicted mortality risk. Findings We included 1488 participants, of whom 236 (15.9%) and 469 (31.5%) met the 2-week and 10-week mortality outcomes, respectively. In the development cohort (n=1263), five variables were selected into the basic model (haemoglobin, neutrophil count, Eastern Cooperative Oncology Group performance status, Glasgow coma scale and treatment regimen), with two additional variables in the research model (cerebrospinal fluid quantitative culture and opening pressure) for 2-week mortality. During internal-external cross-validation, both models showed consistent discrimination across countries (pooled areas under the receiver operating characteristic curves (AUROCs) 0.75 (95% CI 0.68-0.82) and 0.78 (0.75-0.82) for the ‘Basic’ and ‘Research’ 2-week mortality models, respectively), with some variation in calibration between sites. Performance was similar in held-out validation (n=225), with the models demonstrating higher net benefit to inform decision-making than alternative approaches including a pre-existing comparator model. In exploratory analyses, treatment effects varied by predicted mortality risk, with a trend towards lower absolute and relative mortality for a single high-dose liposomal Amphotericin B-based regimen (in comparison to 1-week Amphotericin B deoxycholate plus flucytosine) among lower risk participants in the Ambition-cm trial. Interpretation Both models accurately predict mortality, were generalisable across African trial settings, and have potential to be incorporated into future treatment stratification approaches in low and middle-income settings. Funding MRC, United Kingdom (100504); ANRS, France (ANRS12275); SIDA, Sweden (TRIA2015-1092); Wellcome/MRC/UKAID Joint Global Health Trials (MR/P006922/1); European DCCT Partnership; NIHR, United Kingdom through a Global Health Research Professorship to JNJ (RP-2017-08-ST2-012) and a personal Fellowship to RKG (NIHR302829). Evidence before this study There is an urgent need to improve clinical management for HIV-associated cryptococcal meningitis in resource limited settings across Africa. Cryptococcal meningitis accounts for ∼112,000 AIDS-related deaths per year globally, with over 75% in Africa, despite widespread antiretroviral therapy roll-out. The development of practical tools to identify patients at highest risk of death could help to tailor management strategies and stratify therapy. We searched PubMed for studies published between database inception and Jan 12, 2024, using the terms “cryptococcal meningitis”, “HIV”, “human immunodeficiency virus”, “immunocompromised”, “predict*”, and “model*”, with no language restrictions. Three previous studies, all conducted in China, have developed prognostic models for cryptococcal meningitis mortality. Of these, two used statistical methods while the third used machine learning but focused on persons without HIV only. No studies conducted in Africa, specifically targeting people living with HIV, or using both statistical and machine learning approaches in parallel, were identified. Well-developed and validated tools to predict risk of cryptococcal meningitis mortality and guide treatment stratification are thus lacking for resource limited settings in Africa. Added value of this study To our knowledge, this is the largest study to date to develop and validate prediction models for HIV-associated cryptococcal meningitis mortality. We combined high-quality data from the two largest randomised-controlled clinical trials conducted to date for cryptococcal meningitis treatment, with a total sample size of 1488 participants of whom 236 (15.9%) and 469 (31.5%) met the 2-week and 10-week mortality outcomes, respectively. We developed two models, ‘basic’ and ‘research’, to enable use in both resource-limited and research settings (where additional prognostic markers such as measurements of cerebrospinal fluid (CSF) opening pressure and CSF fungal burden may also be available). In the 2-week mortality models, five variables were included in the ‘basic’ model, with two additional variables included in the ‘research’ model. Both models predicted risk of mortality with consistent discrimination and calibration across sub-Saharan African settings. Head-to-head statistical (logistic regression) and machine learning (XGBoost) methods revealed no added value of the machine learning approach. In exploratory analyses, treatment effects varied by predicted 2-week mortality risk, thus providing proof-of-concept for future treatment stratification approaches. Specifically, there was a trend towards lower mortality for a single high-dose liposomal Amphotericin B-based regimen (in comparison to 1-week Amphotericin B deoxycholate plus flucytosine) among lower risk participants in the Ambition-cm trial. Implications of all the available evidence The personalised risk predictor for cryptococcal meningitis (PERISKOPE-CM) models accurately predicted mortality risk among patients with HIV-associated cryptococcal meningitis and demonstrated generalisable performance across trial settings in Africa. Predictions from the models could be utilised to direct treatment stratification approaches in future clinical trials, with patients at lowest predicted risk receiving less intensive and less toxic therapy. The models have been made available for future research use on an open access online interface. INTRODUCTION Cryptococcal meningitis is a major driver of global HIV-related mortality resulting in ∼112,000 deaths per year, with over 75% in Africa, and accounting for 19% of all AIDS-related deaths. 1 Despite the rapid roll-out of antiretroviral therapy (ART), incidence remains high and ten-week mortality ranges from 24% to over 50%, depending on setting and treatment used. 2 – 7 Stratifying patients by disease severity on admission may allow for more targeted therapy; less sick patients could avoid both prolonged hospitalisation and the most intensive and toxic treatments, whereas those who are severely unwell could be identified early for treatment escalation. Validated approaches to prognostication could be used in interventional trials to direct treatment stratification. Previous studies have identified factors associated with increased risk of HIV-associated cryptococcal meningitis mortality including older age, 8 – 10 clinical measurements at presentation (such as low weight/body mass index (BMI), 8 , 11 , 12 abnormal mental status, 8 , 9 , 12 – 15 and CSF opening pressure 8 , 9 ), and laboratory measurements (including low cerebrospinal fluid (CSF) white cell count (WCC), 12 – 14 , 16 elevated CSF fungal burden, 8 , 10 , 12 , 13 , 15 high peripheral white blood cell (WBC) count, 8 low CD4 cell count, 9 low haemogloblin, 8 and high serum C-reactive protein 17 ). However, the majority of studies published to date are small, with few conducted in Africa. 8 , 11 There are currently no practical tools leveraging the combined discrimination of these factors to identify those at highest risk of death in clinical use. Prognostic models have been developed for a range of acute infectious diseases, notably COVID-19, 18 , 19 where extensive external validation has been performed for a model to predict in-hospital mortality. For cryptococcal meningitis, existing prognostic models have been developed in studies from China using conventional statistical methods 9 , 20 and for persons without HIV using machine learning approaches. 21 However, there are currently no validated models to guide clinical decision-making in people living with advanced HIV in Africa, where the disease burden is highest. In this study, we pooled individual-level data from the two largest randomised-controlled trials (RCTs) to date in HIV-associated cryptococcal meningitis. We aimed to develop and validate prediction models to predict risk of all-cause 2-week mortality. We used both statistical and machine learning approaches, in a head-to-head analysis, and also evaluated performance over a longer (10-week) time horizon. Finally, we assessed whether treatment efficacy in the trials was heterogenous by predicted risk. METHODS Data sources We analysed pooled data from the ACTA ( ISRCTN45035509 ) and Ambition-cm ( ISRCTN72509687 ) phase III, RCTs for HIV-associated cryptococcal meningitis, which have been reported previously. 4 , 5 In brief, adults (≥18 years) with a first episode of HIV-associated cryptococcal meningitis were recruited to both trials. In ACTA, participants were randomized in a 1:1:1 ratio to received either: fluconazole plus flucytosine for 14 days (oral combination regimen); amphotericin B deoxycholate plus either flucytosine or fluconazole for 7 days followed by 7 days fluconazole; or amphotericin B deoxycholate plus either flucytosine or fluconazole for 14 days. Participants were recruited from 9 sites across 4 countries in Africa (Cameroon, Malawi, Tanzania and Zambia). In Ambition-cm, participants were randomized 1:1 to receive either: single high-dose (10 mg/kg) liposomal amphotericin B plus 14 days of flucytosine plus fluconazole (Ambition regimen); or amphotericin B deoxycholate (1 mg/kg/day) plus flucytosine for 7 days, followed by fluconazole for 7 days (1-wk AmBd+5FC regimen). Participants were recruited from 6 sites in 5 countries (Botswana, Malawi, South Africa, Uganda and Zimbabwe). We elected a priori to include all data in model development apart from the Malawi site in the Ambition-cm trial, which was held-out for validation since Malawi was the only country where participants were recruited to both trials. Both studies were approved by the London School of Hygiene and Tropical Medicine Research Ethics Committee and local ethics and regulatory authorities in each country, where appropriate. The full analysis pipeline is summarised in Supplementary Figure 1. All analyses were conducted and reported in accordance with TRIPOD standards 7 and were performed in R (version 4.3.2). Model development We considered predefined candidate predictors for inclusion in the models based on clinical knowledge, previous studies and availability of variables collected at baseline in both trials (Supplementary Table 1). Sample size calculation details are provided in the Supplementary Appendix. Variables were only considered if available from at least 60% of participants, as previously. 23 Missing data were handled using multiple imputation in the primary analyses, using the aRegImpute function in R. 23 , 24 All predictors, including transformations, considered for the final model were included in the imputation model to ensure compatibility. All primary analyses were performed across 10 multiply imputed datasets; parameters were pooled using Rubin’s rules. 25 We developed models for the primary outcome of 2-week mortality, since we hypothesised that early mortality is more likely to be directly associated with cryptococcal meningitis severity. We developed a ‘basic’ model for use in a resource limited setting, using only candidate predictors routinely obtained programmatically at the point of hospital admission; and a ‘research’ model, where all pre-defined candidate predictors were considered for inclusion. Multivariable logistic regression models were initially developed in the development dataset. Variable selection was conducted for the ‘basic’ and ‘research’ models separately, using backward selection, based on Akaike Information Criteria (AIC). Continuous variables were modelled using restricted cubic splines (3 knots) to assess non-linear associations. Predictors retained in >50% of multiply imputed sets were retained in the final models. Models including the selected variables were then evaluated using the internal-external cross-validation method, where participants from one country were iteratively left out of the model development set and used for validation. 23 , 26 , 27 This method evaluates potential generalisability of the modelling approach between settings by examining between-setting heterogeneity in performance. To do this, the models were re-trained in the remaining countries in the development dataset and validated in the omitted country by quantifying the discrimination and calibration. Discrimination assesses how well the models differentiate between those who do and do not meet the outcome (measured as the C-statistic), while calibration evaluates how well predicted risk matches observed risk (measured as the calibration slope and calibration-in-the-large, and by visualisation of calibration plots 28 ). Random-effects meta-analysis was then used to calculate pooled measures of discrimination and calibration across countries in the development dataset. 26 We performed re-calibration by country by re-estimating model intercepts. The final models were trained using the full development dataset prior to further validation in the held-out dataset (Malawi site in the Ambition-cm trial). Machine learning Next, we sought to test whether a statistical model using logistic regression could be improved upon using a machine learning method. We re-trained the research model with the selected variables using XGBoost, as a “best-in-class” machine learning approach for predicting a binary outcome, 29 as previously. 19 Further details of the XGBoost approach are provided in the Supplementary Appendix. Model validation Held-out validation of the statistical and machine learning models was then performed in the Ambition-cm Malawi dataset by quantifying the C-statistic, calibration slope and calibration-in-the-large, and visualisation of calibration plots. We benchmarked performance to single univariable predictors, and other HIV-associated cryptococcal meningitis prognostic models for which constituent variables were available in >60% of the dataset and model reconstruction was possible from reported manuscripts. 9 , 30 We also performed decision curve analysis 30 in the validation dataset to quantify the overall net benefit of implementing the models to inform clinical decisions, compared to: (a) a ‘treat all’ approach; (b) a ‘treat none’ approach; and (c) other candidate models. 10-week analysis Performance of the final logistic regression models to predict mortality over a longer time horizon of 10 weeks was assessed by evaluating discrimination for this outcome in the held-out validation in the Malawi Ambition-cm dataset. In addition, we developed separate models specifically to predict the 10-week outcome, using the same methodology as for the primary models, to assess whether this further improved performance for the longer time horizon. Treatment effect heterogeneity analyses In an exploratory analysis, we also tested the hypothesis that the Ambition regimen and oral combination (fluconazole plus flucytosine) regimen were more effective than 1-wk AmBd+5FC (as the control arm in Ambition-cm and the best performing arm in ACTA), among lower risk participants, in line with the Predictive Approaches to Treatment effect Heterogeneity (PATH) Statement. 31 To do this, we generated predictions for 2-week mortality using the final ‘basic’ and ‘research’ models in the pooled dataset. Participants were divided into low, medium and high-risk strata (based on tertiles of predicted risk). We then examined treatment efficacy of the Ambition and oral combination regimens (compared to 1-wk AmBd+5FC) over 10 weeks separately within each trial, with interaction terms by risk stratum. Absolute risk differences between arms were quantified by risk stratum using an identity link function and relative risk as hazard ratios using Cox regression. In parallel, we evaluated whether treatment efficacy was modified by predicted risk as a continuous variable. We modelled time to death using Cox regression, with an interaction term between treatment arm and predicted 2-week mortality risk (including restricted cubic spline transformations to account for non-linear associations). We then visualised treatment effects (as hazard ratios) against predicted risk. Role of the funding source The funder had no role in study design, data collection, data analysis, data interpretation, writing of the report, or decision to submit for publication. The corresponding authors had full access to all the data in the study and had final responsibility for the decision to submit for publication. RESULTS Baseline characteristics A total of 674 eligible participants from the ACTA trial (excluding 4 participants who were lost to follow-up) and 814 from the Ambition-cm trial were included in the pooled analysis (n=1,488). 1,263 participants were included in model development, with 225 (Malawi Ambition-cm participants) held out for validation (Supplementary Figure 2). Baseline characteristics, including candidate predictors and missingness, are presented in Table 1 , stratified by 2-week mortality. A total of 222/1,263 (17.5%) development set participants and 21/225 (9.3%) validation set participants met the primary model outcome of 2-week mortality. View this table: View inline View popup Table 1: Baseline characteristics of study cohort Characteristics showed both in total study cohort and stratified by two-week mortality outcome. Variable selection In variable selection, five predictors were retained in the ‘basic’ and seven in the ‘research’ model. Predictors in both models included: Glasgow coma scale (GCS) score, Eastern Cooperative Oncology Group (ECOG) performance status, haemoglobin, blood neutrophil count and treatment regimen. Additional predictors in the research model were CSF opening pressure (CSF OP) and log10CSF quantitative cryptococcal culture (CSF QCC). Predictor-outcome associations were similar in both models ( Figure 1 ; Supplementary Figure 3). Full model coefficients for both models are presented in Supplementary Tables 4 and 5. Download figure Open in new tab Figure 1: Multivariable associations between selected predictors and outcome in primary research model Continuous variables were modeled using restricted cubic splines. The final model parameters are pooled across multiply imputed datasets (total sample size for model development = 1,263 participants). For continuous variables, black lines represent point estimates and grey shaded regions represent 95% confidence intervals. For categorical variables, black dots represent point estimates and black lines represent 95% confidence intervals. Treatment arm 1 through 5 represent 1) the liposomal-Amphotericin-B Ambition regimen and the 1-week Amphotericin-B + Flucytosine (1wk AmBd + 5FC) arms from both ACTA and Ambition-cm trials, 2) 1 week Amphotericin-B + Fluconazole, 3) 2 weeks Amphotericin-B + Flucytosine, 4) 2 weeks Amphotericin-B + Fluconazole and 5) Flucytosine + Fluconazole oral combination regimen, respectively. Treatment arm (1) was grouped in this way due to the finding of non-inferiority between the Ambition regimen and 1wk AmBd + 5FC arms in the Ambition-cm trial. Associations for the primary basic model are shown in the appendix (Supp Fig 1). ECOG = Eastern Cooperative Oncology Group performance status; GCS = Glasgow Coma Scale; CSF = cerebrospinal fluid. Internal-external cross-validation Forest plots showing discrimination and calibration metrics of both models from each country in internal-external cross validation in the development dataset are shown in Figure 2 . Discrimination was relatively consistent between study sites for both models (pooled C-statistic estimates 0.75 (95%CI 0.68-0.82) for the ‘basic’ and 0.78 (95%CI 0.75-0.82) for the ‘research’ models). Calibration was more heterogeneous, with CITL varying by study site in both models, likely reflecting variation in baseline risk between sites (point estimates -0.45 to 0.57 for the ‘basic’ model, -0.34 to 0.42 for the ‘research’ model). Pooled calibration plots by study site show evidence of systematic overestimation of risk in Malawi and underestimation in Uganda. Recalibration to each study site by re-estimation of the model intercept led to improvement in model calibration, as expected (Supplementary Figure 4). Download figure Open in new tab Download figure Open in new tab Figure 2 a and 2b: Internal-external cross validation of basic and research model by country Pooled estimates are calculated through random-effects meta-analysis (total sample size = 1,263 participants). Countries with n < 100 participants or x < 20 deaths were amalgamated and grouped by similarity of healthcare environment. Dashed lines indicate lines of perfect calibration in the large (0) and slope (1), respectively. Black squares indicate point estimates; bars indicate 95% confidence intervals; diamonds indicate pooled random-effects meta-analysis estimates. Bots = Botswana; SAfr = South Africa; Tan = Tanzania; Zam = Zambia; Zim = Zimbabwe. Held-out validation In held-out validation in Malawi participants in the Ambition-cm trial, discrimination of both models was slightly higher compared to the estimates in internal-external cross-validation ( Table 2 ) with C-statistic 0.78 [95%CI 0.70-0.87] and 0.85 [95%CI 0.79-0.92] in the ‘basic’ and ‘research’ model, respectively. Calibration assessment suggested overestimation of risk, particularly in the higher risk range, where data were sparser ( Figure 3a-b ). ECOG performance status was the strongest univariable predictor for 2-week mortality (C-statistic 0.78 [95%CI 0.71 to 0.85]) but had lower discrimination than the full multivariable ‘research’ model (Supplementary Table 6). The discriminative ability of single predictors varied by study site (Supplementary Table 7). Of the identified pre-existing models, only the model created by Zhao et al. (2021) 9 could be reconstructed from the available data. Discrimination in the validation dataset, calculated using their reported point score, fell short of either of the primary models (C-statistic 0.69 [95%CI 0.6 to 0.78]). To report calibration, we recalibrated the model intercept to our validation dataset, since no intercept was reported. Despite this, calibration assessment suggested that predictions were too extreme, with a slope 0.35 (95%CI 0.14 - 0.56; Table 2 ; Supplementary Figure 5). Download figure Open in new tab Figure 3: Model calibration and prediction density in held-out validation data Panels (a) to (c) show the calibration of the basic model, research model and XGBoost machine learning model respectively. Calibration is shown using a loess smoother. 95% confidence intervals are shown shaded in grey. Rug plots, shown on the x-axis, plot the distribution of predicted risk. Panels (d) to (f) show density plots for the 2-week mortality predictions made by the basic model, research model and XGBoost machine learning model respectively, stratified by 2-week mortality outcome. View this table: View inline View popup Download powerpoint Table 2: Model performance in held-out validation data Table reports performance in validation data for each of the primary models, secondary models, XGBoost machine learning model and Zhao et al’s model. Brackets in the table headers refer to the outcome for which the model was trained. Validation outcome refers to the outcome in which model performance was assessed. Slope and Calibration-in-the-large data are not reported for the primary models’ 10-week validation outcome, since the primary model was developed for the 2-week mortality outcome. Calibration-in-the-large data is not reported for Zhao et al’s model as in order to reconstruct the model and assess calibration, the model intercept was recalibrated to our data. Machine learning We also developed an XGBoost model, to assess whether a machine learning approach could further improve performance. Predictor-outcome associations using XGBoost were similar to the logistic regression approach, with no clear evidence of two-way interactions (Supplementary Figure 7). In the validation dataset, XGBoost discrimination and calibration metrics were also similar to the logistic regression model ( Table 2 ; Figure 3 ). The XGBoost calibration plot showed miscalibration with predictions being too extreme, reflecting overfitting, and leading to a slope <1 ( Figure 3c ). Overall, performance of the machine learning model demonstrated no improvement in performance over the logistic regression approach. Decision curve analysis Decision curve analysis measures the trade-off between correctly identifying participants who met the mortality endpoint and incorrectly identifying those who did not and quantifies this as “net benefit” 30 . Decision-curve analysis in the validation dataset showed higher net-benefit for the ‘basic’ and ‘research’ models compared with the model by Zhao et al. (2021), 9 ‘treat-all’ or ‘treat-none’ approaches across a broad range of threshold probabilities, where the weighting of the false positives is varied. The ‘research’ model demonstrated superior net benefit overall to guide management (Supplementary Figure 6). 10-week mortality A total of 411/1263 (32.5%) and 58/225 (26.2%) participants met the 10-week mortality outcome in the development and validation datasets, respectively. We retrained the logistic regression models, including variable selection, to assess whether this would improve prediction over the longer time period. Seven predictors were retained in the retrained ‘basic’, and eight in the ‘research’ 10-week models. In the ‘basic’ model, age, weight, and the presence of seizures on admission were substituted for GCS, and in the ‘research’ model age and CSF cell count were substituted for CSF OP (Supplementary Tables 8-10; Supplementary Figure 9). Performance of the 10-week models in the validation dataset is shown in Table 2 , along with the C-statistic for the primary 2-week models in predicting 10-week mortality. Overall, for both ‘basic’ and ‘research’ models, discrimination for the 10-week outcome was similar for the re-trained 10-week models and the primary 2-week models. For both the ‘basic’ and ‘research’ 10-week models, the CITL was closer to 0 for the 10-week outcome, when compared to the primary 2-week models for the 2-week outcome (Supplementary Figure 10), suggesting better calibration for the 10-week models. Treatment effect modification analysis We hypothesised that treatment effects within each trial may vary according to predicted 2-week mortality risk. Predicted risk for both models was positively skewed, with modal risk of 7.6% and 4.2% for the ‘basic’ and ‘research’ models, respectively, markedly lower than the overall cohort mortality of 16.3% (Supplementary Figure 11). Absolute number of deaths, mortality risk difference and hazard ratios comparing the Ambition regimen to 1-wk AmBd+5FC in the Ambition-cm trial and oral combination treatment to 1-wk AmBd+5FC in the ACTA trial, stratified by risk stratum, are shown in Supplementary Table 11. In the Ambition-cm trial, there was a trend towards absolute and relative risk reductions for the Ambition regimen for the lower risk strata for both models, with no difference in mortality in the higher risk stratum: Hazard ratio 0.65 (95%CI 0.38-1.13) for ‘basic’ model in low-risk stratum, compared to 1.05 (95%CI 0.71-1.54) in high-risk stratum; 0.61 (95%CI 0.33-1.11) for ‘research’ model in low-risk stratum, compared to 1 (95%CI 0.68-1.45) in high-risk stratum). When comparing oral combination treatment to 1-wk AmBd+5FC in the ACTA trial, risk differences and hazard ratios appeared closer to 1 in the low- and medium-risk strata, compared to the high-risk strata, for both models, though the confidence intervals were wide due to small sample sizes in each stratum. These findings were similar when handling predicted risk as a continuous variable, shown in Figure 4 . Download figure Open in new tab Figure 4: Treatment Effect of Oral regimen (flucytosine (5FC) plus fluconazole (FLU)) and Ambition regimen vs 1-week Amphotericin B deoxycholate (AmBd) plus 5FC over a range of treatment model-predicted mortality Treatment effect, expressed as a hazard ratio for the Oral or Ambition regimens relative to 1-week of Amphotericin B plus Flucytosine (1wk AmBd + 5FC), is plotted logarithmically to base 2 on the y axis. Predicted mortality, defined by either the Basic or Research model as per the panel labels, is plotted on the x axis. Green, Orange and Red colouring represent the Low, Medium and High Risk terciles defined by the research model. Shaded regions represent 95% confidence intervals. The x-axis is foreshortened to predicted mortality of 0.4, which encompasses >95% of the underlying data. DISCUSSION Using a large, high-quality dataset of the 2 largest randomised controlled trials conducted to date for the treatment of cryptococcal meningitis, conducted across 8 countries in Africa, we developed tools to predict 2-week and 10-week mortality. Our ‘basic’ model includes five predictor variables, which are routinely measured programmatically: GCS score, ECOG performance status, haemoglobin, neutrophil count and treatment regimen. Our ‘research’ model additionally includes CSF OP and CSF fungal burden, for settings where these variables are available. Both models demonstrated consistent performance across countries, supporting their generalisability. Predicted risk was markedly skewed with modal mortality risk far below overall risk, suggesting that the majority of patients have individual-level risk far below the mean and thus supporting an individualised approach to treatment. Moreover, our results provide proof of concept data for treatment stratification. While the Ambition regimen was non-inferior to 1-wk AmBd+5FC overall in the Ambition-cm trial, we observed a trend towards improved mortality with the Ambition regimen among lower risk participants, when using both categorical and continuous risk measurements. Treatment regimens in cryptococcal meningitis are intensive and use of Amphotericin B deoxycholate is associated with significant toxicity risk. 32 One potential hypothesis is that treatment toxicity may be a relatively more important factor determining outcome among lower-risk participants, which may explain the apparent improved outcomes with the less toxic Ambition regimen 5 among these participants. When evaluating oral combination therapy in the ACTA trial, the oral regimen was close to the efficacy of 1-wk AmBd+5FC in the lowest risk participants. Our results suggest our models could be used to direct stratified treatment approaches in future trials, with patients at lowest predicted risk receiving less toxic or intensive therapy and being considered for earlier hospital discharge. Encouragingly, this effect was seen in both models, suggesting the ‘basic’ version of the model could be used if measurement of CSF QCC and opening pressure is not possible. The models demonstrated consistent discrimination across different study sites but demonstrated some variation in calibration, likely reflecting site-level differences in baseline mortality risk not accounted for in the model. In Malawi, the model demonstrated systematic overprediction of risk and in Uganda underprediction, indicating that patients had a less-than and greater-than expected mortality at these sites, respectively. Multiple explanations are plausible, including differences in healthcare provision, genetic predisposition to outcome, and other local socio-economic determinants. Recalibration of the models to individual study sites resolved much of this difference and represents a possible way to adapt the models in future, if required. Nevertheless, in the held-out validation data, discrimination and calibration performance of both ‘basic’ and ‘research’ models improved significantly on the most comparable existing cryptococcal meningitis model for people living with HIV 9 and is consistent with that achieved to predict mortality in COVID-19. 23 The final models demonstrated higher net benefit than alternative approaches to inform decision making in decision curve analysis, without recalibration. Notably, most of the observed miscalibration was at higher predicted risks (>25%), where the data were sparser. However, since this represents a risk range above the likely threshold probability for most stratified interventions, this is unlikely to have a significant impact on clinical utility. Our analyses demonstrated clinically plausible predictor-outcome associations for all included variables in the models and explored non-linear associations of continuous variables with mortality. GCS score <15 and elevated CSF fungal burden have been found to be associated with increased mortality in previous studies from multiple settings. 8 , 9 , 12 – 15 , 20 , 21 , 33 Notably, CSF fungal burden at baseline, measured using quantitative cultures, had a non-linear association with mortality risk with markedly increased risk above log104, but little increased risk of mortality at lower levels. Although measurement of CSF fungal burden using quantitative cultures is not a routine laboratory procedure, it may be possible in future to utilise results from point-of-care semi-quantitative cryptococcal antigen tests or other approaches to quantify organism load such as qPCR. 34 We found that ECOG performance status, reflecting functional status, was the most discriminating single predictor for mortality. Since it is easy to measure at presentation, there is a strong rationale for considering this both in research studies and programmatically. Raised intracranial pressure (ICP) is well-known to contribute to morbidity and mortality in HIV-associated cryptococcal meningitis 9 , 35 , 36 and is common on presentation with 580/1488 (39.0%) in the current study having ICP >25cm H2O. The risk associated with 2-week mortality increased modestly with increasing OPs in our analysis. Since raised ICP was managed through protocolised therapeutic lumbar puncture in our included trials, the effect of unmanaged raised ICP on mortality was likely attenuated, as has been hypothesised in previous analyses. 8 Thus, the association observed in our models should be considered to reflect ICP with protocolised management. Anaemia and high blood neutrophil count were also associated with increased 2-week mortality in our models, as observed for other infectious diseases, and likely reflecting severity of systemic illness. 8 In our re-trained models for 10-week mortality, age, weight, and seizures on admission were included as additional variables in the ‘basic’ model, while age and CSF cell count were substituted for CSF OP in the ‘research’ model. We hypothesise that most 2-week mortality risk likely relates to cryptococcal meningitis-related pathology, whereas at 10 weeks a larger proportion of the observed risk may be attributable to the underlying HIV and co-morbidities. Since 10-week mortality is significantly higher than 2-week mortality, there were also a larger number of events for the 10-week outcome, which may have led to the inclusion of more variables and slightly better calibration observed for these models. Though previous studies have sought to compare performance of statistical and machine learning prediction models, most comparisons were classified at high risk of bias in a recent systematic review, due to suboptimal statistical and/or machine learning methodology. 37 We found that performance of the machine learning model in the validation dataset demonstrated no improvement in performance over the logistic regression model. Our findings suggest that, in lower dimensional datasets, traditional regression approaches may offer equivalent performance whilst remaining computationally less intensive and methodologically more transparent. Notably, our observed predictor-outcome associations from the XGBoost approach largely mirrored the associations found in the regression model. Strengths of our study include our use of best practices for statistical and machine learning prediction modelling development and validation, including TRIPOD standard reporting, using multiple imputation to deal with missing data, and retaining continuous variables without arbitrary categorisation to avoid loss of information, whilst also accounting for non-linear associations. We also used the largest dataset to date to develop and validate a model for cryptococcal meningitis and mitigated the risk of overfitting our models by defining candidate predictors a prior i, in line with best practice sample size guidance. 38 Furthermore, our pooled dataset included 8 countries; we harnessed this to explore generalisability and geographic heterogeneity in model performance through our internal-external cross-validation approach. Finally, we explored heterogeneity of treatment effect by predicted risk, in line with the PATH statement. 31 Our study also has limitations. First, our models were developed and validated using data from sub-Saharan Africa, where the greatest burden of disease is; evaluation in other world regions is required to further explore generalisability. While developing the model on RCT data was a strength in terms of data quality, real-world outcomes for cryptococcal meningitis may be inferior to that observed in the trials, due to factors such as trial exclusion criteria and improved standards of care in clinical trials. 39 Further validation in programmatic cohorts is therefore also required. Whilst our treatment effect analysis provides early proof-of-concept evidence for treatment stratification in cryptococcal meningitis, caution is required due to the exploratory nature of the analysis. While the models were not trained to predict differential treatment effects, their discrimination and calibration to predict 2-week mortality is likely optimistic across participants in the development dataset. Future studies are required to further evaluate differential treatment effects when stratified by our models, including RCTs incorporating approaches to treatment stratification. In summary, we present prognostic models for 2- and 10-week mortality in HIV-associated cryptococcal meningitis and demonstrate consistent performance across sub-Saharan African settings. The models use commonly available predictors and will be made freely available to direct future treatment stratification approaches in clinical trials (prototype: https://rishi-k-gupta.shinyapps.io/periskope-cm/ ). Author contributions SFM, TSH, JNJ and RKG conceived the study. THAS, SFM, TSH, JNJ and RKG wrote the analysis plan. SFM curated the pooled dataset. THAS and RKG performed the analysis. THAS, SFM and RKG wrote the first draft of the manuscript, with input from TSH and JNJ. RKG built the Rshiny app and WSL developed the PERISKOPE website. All contributing authors reviewed and approved the manuscript prior to submission. Funding The contributing primary trials were funded by the MRC, United Kingdom (100504); ANRS, France (ANRS12275); SIDA, Sweden (TRIA2015-1092); Wellcome/MRC/UKAID Joint Global Health Trials (MR/P006922/1); and European DCCT Partnership. JNJ is funded by an NIHR Global Health Research Professorship (RP-2017-08-ST2-012). RKG is funded by the NIHR (NIHR302829, NIHR303184) and by NIHR Biomedical Research Funding to UCL and UCLH. Declaration of interests DSL has received salary support from Janssen. The authors have no other declarations of interest. Data availability statement The individual-level data for the trials included in this pooled analysis will be made available to investigators upon reasonable request to the corresponding authors of the original trials. Acknowledgements We would like to acknowledge all site staff and participants contributing to the primary randomised-controlled trials. Footnotes Table of authors: Thomas HA Samuels Dr. MRCP, Sile F Molloy Dr. Ph.D., David S Lawrence Dr. Ph.D., Angela Loyse Dr. M.D.(Res.), Cecilia Kanyama Dr. M.D., Robert S Heyderman Professor FMedSci, Wai Shing Lai Mr. BSc, Sayoki Mfinanga Professor Ph.D., Sokoine Lesikari Dr M.D., Duncan Chanda Dr. Ph.D., Charles Kouanfack Dr. Ph.D., Elvis Temfack Dr. M.D., Olivier Lortholary Professor Ph.D., Mina C Hosseinipour Professor M.D., Adrienne K Chan Dr. M.D., David B Meya Dr. Ph.D., David R Boulware Professor M.D., Henry C Mwandumba Professor Ph.D., Graeme Meintjes Professor Ph.D., Conrad Muzoora Dr. M.D., Mosepele Mosepele Professor. M.D., Chiratidzo E Ndhlovu Professor F.R.C.P., Nabila Youssouf Dr. Ph.D., Thomas S Harrison Professor M.D., Joseph N Jarvis Professor Ph.D., Rishi K Gupta Dr. Ph.D. REFERENCES 1. ↵ Rajasingham R , Govender NP , Jordan A , et al. The global burden of HIV-associated cryptococcal infection in adults in 2020: a modelling analysis . Lancet Infect Dis 2022 ; 22 ( 12 ): 1748 – 55 . OpenUrl 2. ↵ Mwaba P , Mwansa J , Chintu C , et al. Clinical presentation, natural history, and cumulative death rates of 230 adults with primary cryptococcal meningitis in Zambian AIDS patients treated under local conditions . Postgraduate medical journal 2001 ; 77 ( 914 ): 769 – 73 . OpenUrl Abstract / FREE Full Text 3. Longley N , Muzoora C , Taseera K , et al. Dose response effect of high-dose fluconazole for HIV-associated cryptococcal meningitis in southwestern Uganda . Clin Infect Dis 2008 ; 47 ( 12 ): 1556 – 61 . OpenUrl CrossRef PubMed Web of Science 4. ↵ Molloy SF , Kanyama C , Heyderman RS , et al. Antifungal Combinations for Treatment of Cryptococcal Meningitis in Africa . New England Journal of Medicine 2018 ; 378 ( 11 ): 1004 – 17 . OpenUrl CrossRef PubMed 5. ↵ Jarvis JN , Lawrence DS , Meya DB , et al. Single-Dose Liposomal Amphotericin B Treatment for Cryptococcal Meningitis . New England Journal of Medicine 2022 ; 386 ( 12 ): 1109 – 20 . OpenUrl CrossRef 6. Gaskell KM , Rothe C , Gnanadurai R , et al. A Prospective Study of Mortality from Cryptococcal Meningitis following Treatment Induction with 1200mg Oral Fluconazole in Blantyre, Malawi . PLoS One 2014 ; 9 ( 11 ): e110285 . OpenUrl CrossRef PubMed 7. ↵ Jackson AT , Nussbaum JC , Phulusa J , et al. A phase II randomized controlled trial adding oral flucytosine to high-dose fluconazole, with short-course amphotericin B, for cryptococcal meningitis . AIDS 2012 ; 26 ( 11 ): 1363 – 70 . OpenUrl CrossRef PubMed Web of Science 8. ↵ Jarvis JN , Bicanic T , Loyse A , et al. Determinants of mortality in a combined cohort of 501 patients with HIV-associated Cryptococcal meningitis: implications for improving outcomes . Clin Infect Dis 2014 ; 58 ( 5 ): 736 – 45 . OpenUrl CrossRef PubMed 9. ↵ Zhao T , Xu XL , Nie JM , et al. Establishment of a novel scoring model for mortality risk prediction in HIV-infected patients with cryptococcal meningitis . BMC Infect Dis 2021 ; 21 ( 1 ): 786 . OpenUrl 10. ↵ de Oliveira L , Melhem MdSC , Buccheri R , Chagas OJ , Vidal JE , Diaz-Quijano FA . Early clinical and microbiological predictors of outcome in hospitalized patients with cryptococcal meningitis . BMC Infectious Diseases 2022 ; 22 ( 1 ): 138 . OpenUrl 11. ↵ Kendi C , Penner J , Koech J , et al. Predictors of outcome in routine care for Cryptococcal meningitis in Western Kenya: lessons for HIV outpatient care in resource-limited settings . Postgraduate medical journal 2013 ; 89 ( 1048 ): 73 – 7 . OpenUrl Abstract / FREE Full Text 12. ↵ Anekthananon T , Manosuthi W , Chetchotisakd P , et al. Predictors of poor clinical outcome of cryptococcal meningitis in HIV-infected patients . Int J STD AIDS 2011 ; 22 ( 11 ): 665 – 70 . OpenUrl CrossRef PubMed 13. ↵ Saag MS , Powderly WG , Cloud GA , et al. Comparison of amphotericin B with fluconazole in the treatment of acute AIDS-associated cryptococcal meningitis. The NIAID Mycoses Study Group and the AIDS Clinical Trials Group . The New England journal of medicine 1992 ; 326 ( 2 ): 83 – 9 . OpenUrl CrossRef PubMed Web of Science 14. ↵ Dismukes WE , Cloud G , Gallis HA , et al. Treatment of cryptococcal meningitis with combination amphotericin B and flucytosine for four as compared with six weeks . The New England journal of medicine 1987 ; 317 ( 6 ): 334 – 41 . OpenUrl CrossRef PubMed Web of Science 15. ↵ Day JN , Chau TT , Wolbers M , et al. Combination antifungal therapy for cryptococcal meningitis . The New England journal of medicine 2013 ; 368 ( 14 ): 1291 – 302 . OpenUrl CrossRef PubMed Web of Science 16. ↵ Qu J , Jiang J , Lv X . The utility of cerebrospinal fluid white cell count during the prognostic assessment for cryptococcal meningitis patients: a retrospective study . BMC Infect Dis 2020 ; 20 ( 1 ): 571 . OpenUrl 17. ↵ Chesdachai S , Engen NW , Rhein J , et al. Baseline Serum C-Reactive Protein Level Predicts Mortality in Cryptococcal Meningitis . Open Forum Infect Dis 2020 ; 7 ( 12 ): ofaa530. 18. ↵ Jong VMTd, Rousset RZ, Antonio-Villa NE, et al. Clinical prediction models for mortality in patients with covid-19: external validation and individual participant data meta-analysis . BMJ 2022 ; 378 : e069881 . OpenUrl Abstract / FREE Full Text 19. ↵ Knight SR , Ho A , Pius R , et al. Risk stratification of patients admitted to hospital with covid-19 using the ISARIC WHO Clinical Characterisation Protocol: development and validation of the 4C Mortality Score . BMJ 2020 ; 370 : m3339 . OpenUrl Abstract / FREE Full Text 20. ↵ Zhang C , He Z , Tan Z , Tian F . The clinic-based predictive modeling for prognosis of patients with cryptococcal meningitis . BMC Infect Dis 2023 ; 23 ( 1 ): 352 . OpenUrl 21. ↵ Liu J , Lu Y , Liu J , et al. Development and validation of a machine learning model to predict prognosis in HIV-negative cryptococcal meningitis patients: a multicenter study . Eur J Clin Microbiol Infect Dis 2023 ; 42 ( 10 ): 1183 – 94 . OpenUrl 22. World Health Organization . Guidelines for diagnosing, preventing and managing cryptococcal disease among adults, adolescents and children living with HIV . ISBN 978-92-4-005217-8 2022. 23. ↵ Gupta RK , Harrison EM , Ho A , et al. Development and validation of the ISARIC 4C Deterioration model for adults hospitalised with COVID-19: a prospective cohort study . The Lancet Respiratory Medicine 2021 ; 9 ( 4 ): 349 – 59 . OpenUrl 24. ↵ R Core Team . R: A language and environment for statistical computing . R Foundation for Statistical Computing 2021 : Vienna, Austria . https://www.R-project.org/ . 25. ↵ Rubin DB . Multiple Imputation for Nonresponse in Surveys . Hoboken : John Wiley & Sons ; 2004 . 26. ↵ Debray TPA , Riley RD , Rovers MM , Reitsma JB , Moons KGM , Cochrane IPDM -aMg. Individual Participant Data (IPD) Meta-analyses of Diagnostic and Prognostic Modeling Studies: Guidance on Their Use . PLOS Medicine 2015 ; 12 ( 10 ): e1001886 . OpenUrl 27. ↵ Gupta RK , Calderwood CJ , Yavlinsky A , et al. Discovery and validation of a personalized risk predictor for incident tuberculosis in low transmission settings . Nature medicine 2020 ; 26 ( 12 ): 1941 – 9 . OpenUrl PubMed 28. ↵ Steyerberg EW , Vergouwe Y . Towards better clinical prediction models: seven steps for development and an ABCD for validation . European heart journal 2014 ; 35 ( 29 ): 1925 – 31 . OpenUrl CrossRef PubMed Web of Science 29. ↵ Chen T , He T , Benesty M , et al. _xgboost: Extreme Gradient Boosting_. R package version 1.7.6.1 . 2023 . 30. ↵ Vickers AJ , Van Calster B , Steyerberg EW . Net benefit approaches to the evaluation of prediction models, molecular markers, and diagnostic tests . Bmj 2016 ; 352 : i6 . OpenUrl FREE Full Text 31. ↵ Kent DM , Paulus JK , van Klaveren D , et al. The Predictive Approaches to Treatment effect Heterogeneity (PATH) Statement . Annals of Internal Medicine 2019 ; 172 ( 1 ): 35 – 45 . OpenUrl CrossRef 32. ↵ Bicanic T , Bottomley C , Loyse A , et al. Toxicity of Amphotericin B Deoxycholate-Based Induction Therapy in Patients with HIV-Associated Cryptococcal Meningitis . Antimicrob Agents Chemother 2015 ; 59 ( 12 ): 7224 – 31 . OpenUrl Abstract / FREE Full Text 33. ↵ Majumder S , Mandal SK , Bandyopadhyay D . Prognostic markers in AIDS-related cryptococcal meningitis . J Assoc Physicians India 2011 ; 59 : 152 – 4 . OpenUrl PubMed 34. ↵ Mbangiwa T , Sturny-Leclère A , Lechiile K , et al. Development and validation of quantitative PCR assays for HIV-associated cryptococcal meningitis in sub-Saharan Africa: a diagnostic accuracy study . The Lancet Microbe 2024 ; 5 ( 3 ): e261 – e71 . OpenUrl 35. ↵ Graybill JR , Sobel J , Saag M , et al. Diagnosis and management of increased intracranial pressure in patients with AIDS and cryptococcal meningitis. The NIAID Mycoses Study Group and AIDS Cooperative Treatment Groups . Clin Infect Dis 2000 ; 30 ( 1 ): 47 – 54 . OpenUrl CrossRef PubMed Web of Science 36. ↵ van der Horst CM , Saag MS , Cloud GA , et al. Treatment of cryptococcal meningitis associated with the acquired immunodeficiency syndrome. National Institute of Allergy and Infectious Diseases Mycoses Study Group and AIDS Clinical Trials Group . The New England journal of medicine 1997 ; 337 ( 1 ): 15 – 21 . OpenUrl CrossRef PubMed Web of Science 37. ↵ Christodoulou E , Ma J , Collins GS , Steyerberg EW , Verbakel JY , Van Calster B . A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models . Journal of clinical epidemiology 2019 ; 110 : 12 – 22 . OpenUrl CrossRef PubMed 38. ↵ Riley RD , Ensor J , Snell KIE , et al. Calculating the sample size required for developing a clinical prediction model . Bmj 2020 ; 368 : m441 . OpenUrl FREE Full Text 39. ↵ Lawrence DS , Leeme T , Mosepele M , Harrison TS , Seeley J , Jarvis JN . Equity in clinical trials for HIV-associated cryptococcal meningitis: A systematic review of global representation and inclusion of patients and researchers . PLoS neglected tropical diseases 2021 ; 15 ( 5 ): e0009376 . OpenUrl View the discussion thread. Back to top Previous Next Posted July 10, 2024. Download PDF Supplementary Material Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Personalised risk prediction tools for cryptococcal meningitis mortality to guide treatment stratification; a pooled analysis of two randomised-controlled trials Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Personalised risk prediction tools for cryptococcal meningitis mortality to guide treatment stratification; a pooled analysis of two randomised-controlled trials T.H.A. Samuels , S.F. Molloy , D.S. Lawrence , A. Loyse , C. Kanyama , R.S. Heyderman , W.S. Lai , S. Mfinanga , S. Lesikari , D. Chanda , C. Kouanfack , E. Temfack , O. Lortholary , M.C. Hosseinipour , A.K. Chan , D.B. Meya , D.R. Boulware , H.C. Mwandumba , G. Meintjes , C. Muzoora , M. Mosepele , C.E. Ndhlovu , N. Youssouf , T.S. Harrison , J.N. Jarvis , R.K. Gupta medRxiv 2024.07.10.24310212; doi: https://doi.org/10.1101/2024.07.10.24310212 Share This Article: Copy Citation Tools Personalised risk prediction tools for cryptococcal meningitis mortality to guide treatment stratification; a pooled analysis of two randomised-controlled trials T.H.A. Samuels , S.F. Molloy , D.S. Lawrence , A. Loyse , C. Kanyama , R.S. Heyderman , W.S. Lai , S. Mfinanga , S. Lesikari , D. Chanda , C. Kouanfack , E. Temfack , O. Lortholary , M.C. Hosseinipour , A.K. Chan , D.B. Meya , D.R. Boulware , H.C. Mwandumba , G. Meintjes , C. Muzoora , M. Mosepele , C.E. Ndhlovu , N. Youssouf , T.S. Harrison , J.N. Jarvis , R.K. Gupta medRxiv 2024.07.10.24310212; doi: https://doi.org/10.1101/2024.07.10.24310212 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Infectious Diseases (except HIV/AIDS) Subject Areas All Articles Addiction Medicine (574) Allergy and Immunology (865) Anesthesia (304) Cardiovascular Medicine (4462) Dentistry and Oral Medicine (445) Dermatology (383) Emergency Medicine (611) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1517) Epidemiology (15251) Forensic Medicine (31) Gastroenterology (1132) Genetic and Genomic Medicine (6621) Geriatric Medicine (669) Health Economics (1002) Health Informatics (4564) Health Policy (1372) Health Systems and Quality Improvement (1617) Hematology (544) HIV/AIDS (1272) Infectious Diseases (except HIV/AIDS) (15938) Intensive Care and Critical Care Medicine (1107) Medical Education (624) Medical Ethics (147) Nephrology (670) Neurology (6643) Nursing (346) Nutrition (1001) Obstetrics and Gynecology (1149) Occupational and Environmental Health (957) Oncology (3350) Ophthalmology (981) Orthopedics (369) Otolaryngology (421) Pain Medicine (436) Palliative Medicine (130) Pathology (665) Pediatrics (1698) Pharmacology and Therapeutics (694) Primary Care Research (714) Psychiatry and Clinical Psychology (5465) Public and Global Health (9259) Radiology and Imaging (2212) Rehabilitation Medicine and Physical Therapy (1372) Respiratory Medicine (1199) Rheumatology (598) Sexual and Reproductive Health (716) Sports Medicine (533) Surgery (715) Toxicology (100) Transplantation (289) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'a03e2dc47bc64193',t:'MTc4MDE0ODcyMQ=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();