Long-term Adherence Assessment Of People Living With HIV Based On Their Hair Concentrations Of Multiple Antiretroviral Drugs

Nowadays,the antiretroviral therapy(ART),consisting of multiple antiretroviral(ARV)drugs,is the cornerstone for HIV prevention and treatment.The long-term adherence to ART is the key to achieve positive treatment outcomes for people living with HIV(PLWH).Therefore,an assessment of long-term ART adherence is vital to improve clinical treatment outcomes and avoid drug resistance among PLWH.In recent years,hair ARV concentrations as an innovative and non-invasive biomarker for measuring cumulative ARV drugs’ exposure has received increasing research interest in assessing long-term adherence and further predicting clinical treatment outcomes of PLWH.However,there are still some challenges.Firstly,the long-term stability of hair ARV concentrations is the key issue for accurately assessing the adherence during a long-time period,but there was no study that systematically explored the issue.Secondly,existing studies mostly utilized the hair concentrations of a single ARV drug or several target drugs from an identical regimen rather than all kinds of drugs to assess ART adherence.Actually,PLWH might often miss drug doses and/or some kinds of drugs from an identical regimen or administer wrong drugs because of adverse side effect and the changes of ART regimen.Accordingly,the hair concentrations of a single drug or several target drugs may only reflect the real administration of the determined drugs,rather than the other drugs from an identical regimen.Thirdly,in order to examine the reliability of hair ARV concentrations in adherence assessment,most studies estimated its prediction performance on virological outcomes rather than immunological outcomes.Fourthly,few previous studies used a large-scale PLWH cohort to calculate the optimal classification thresholds of hair ARV concentrations for predicting clinical treatment outcomes or screen the optimal machine-learning prediction model for accurate diagnosis of clinical outcomes.Currently,there are six mainstream ARV drugs(i.e.,lamivudine,zidovudine,nevirapine,efavirenz,ritonavir,and lopinavir)that recommended by the National Free Antiretroviral Treatment Program for most Chinese PLWH on free ART treatment.Overall,this study aimed to develop a novel measure in the term of the hair concentrations of the aforementioned six ARV drugs to accurately assess long-term adherence of Chinese PLWH and further predict their clinical treatment outcomes.In this study,a method based on high performance liquid chromatography tandem mass spectrometry(LC-MS/MS)was firstly developed for simultaneous determination of the six ARV drugs in hair from PLWH.Thereafter,data from two large-scale PLWH cohorts(cohorts A and B)were utilized to systematically explore the effectiveness,reliability and clinical applicability of the aforementioned hair ARV concentrations through estimating its consistency with traditional self-reported measures and long-term stability across one year,and verifying its prediction performances on virological and immunological outcomes,and screening its optimal classification threshold and machine-learning model.The main results were expressed as follows:Firstly,the six ARV concentrations in 10-mg hair from PLWH were simultaneously determined with the newly-developed LC-MS/MS method where a mixture of methanol and 4m M ammonium acetate in tri-distilled water(95:5,v/v)was used as mobile phase and electronic spray ionization operating in multiple reactions monitoring positive mode as mass-spectromic source.The present method showed good performances with limits of quantitation from 6 to 12 pg/mg,inter-and intra-day coefficients of variation <12.4%,and the recovery ranging at 91.1-113.7%.Moreover,other validation parameters(i.e.,selectivity,matrix effect,stability and carry-over)also met the acceptance criteria of FDA and EMA guidelines.Secondly,this study utilized a three-wave longitudinal design across one year(time interval=6 months)where the 1-cm hair segment closest to the scalp and six self-reported measures from 304 PLWH(cohort A)were collected for three times.Results revealed that hair concentrations of lamivudine,efavirenz,ritonavir and lopinavir all showed significant correlations with at least one of six self-reported measures in the average levels from the three-time measures(rs=0.138-0.513,ps<0.028),indicating that there were weak to moderate long-term consistencies between the two biomarkers.Thirdly,this study utilized the same three-wave longitudinal design on cohort A for long-term stability examination.Results revealed that the hair concentrations of the six ARV drugs all had moderate and even strong long-term relative stability,showing significant Pearson’s correlations(rs=0.375-0.827,ps<0.001)and intra-class correlations(ICCs=0.750-0.901,ps<0.001).Moreover,the hair concentrations of nevirapine,ritonavir and lopinavir also exhibited long-term absolute stability,showing no differences across three-time measures(ps>0.369).In contrast,all self-reported measures showed no long-term relative stability(ps>0.076),but had absolute stability(ps>0.062).Additionally,the hair ARV concentrations showed stronger long-term internal correlations than self-reported measures(rs=0.478-0.942 vs 0.146-0.744).Fourthly,a cross-sectional design based on 1293 PLWH(cohort B)was conducted.The1-cm hair segment closest to the scalp and two self-reported measures were collected once.The receiver operating characteristics(ROC)curves were drawn,and the binary logistic regressions with multivariate models were performed and the machine-learning prediction models with three different algorithms(i.e.,random forests,support vector machine and neural networks)were established to verify the performance of hair ARV concentrations in predicting clinical treatment outcomes.ROC curves revealed that the hair concentrations of the six ARV drugs all had moderate and stronger performances in predicting virological outcomes(area under the ROC curves(AROCs)=0.633-0.833,ps<0.05).Furthermore,after adjusting the socio-demographics and clinical characteristics that correlated with virological outcomes and hair ARV concentrations,in the binary logistic regressions with multivariate models based on the optimal classification thresholds calculated with ROC curves,six hair biomarkers all remained the strong prediction performances [adjusted odds ratios(a ORs)>4.052,ps<0.001],which were better than those from two self-reported measures.However,only hair nevirapine concnetrations showed moderate performance(AROC=0.565,p=0.063 and a OR=2.143,p=0.001)but other five hair biomarkers showed weak performances that in line with those from two self-reported measures in predicting immunological outcomes.Similarly,the machine learning-based univariate prediction model with the three algorithms also revealed that hair ARV concentrations showed stronger prediction performances on virological outcomes(accuracy=59.65-95.46% vs 52.75-93.96%)and similar performances on immunological outcomes(accuracy=49.05-58.93% vs 47.14-58.50%)in comparing with self-reported measures.Furthermore,the multivariate prediction models including adherence as predictor showed improved performance than that only including soci-demographics and HIV-related clinical variates for predicting virological outcomes and immunological outcomes.Especially,the models using hair ARV concentrations as adherence measures exhibited higher accuracy than that using self-reported measures with the improved increase over 110%.Finally,the optimal classification thresholds of the hair ARV concentrations were calculated and verified at 269 pg/mg for lamivudine,58 pg/mg for zidovudine,11415 pg/mg for nevaripine,3094 pg/mg for efavirenz,216 pg/mg for ritonavir and 1431 pg/mg for lopinavir using the ROC curves and binary logistic regression models,which exhibited good clinical applicability in predicting virological outcomes among PLWH.Moreover,the optimal classification threshold was 12624 pg/mg for hair nevirapine concentrations in the prediction of immunological outcomes.In summary,the present novel measure consisting of hair concentrations of six ARV drugs(i.e.,lamivudine,zidovudine,nevirapine,efavirenz,ritonavir and lopinavir)could have stronger long-term stability and higher accuracy in predicting clinical treatment outcomes than self-reported measures,therefore showing better performances in comprehensively and accurately assessing long-term adherence among most Chinese PLWH and further predicting their clinical outcomes to avoid ART treatment failure.