Introduction: Even if there are different barriers to HIV testing in developing countries, timely access to care and early diagnosis improves the clinical course of the disease, reduces the transmission rates, and prolongs the survival time of patients. 

Objectives: This study aimed to investigate the potential predictors of survival time in a cohort of HIV-infected patients on ART using the parametric survival model perspective.

Data and Methods: A hospital-based retrospective cohort charts-review study design was conducted on 647 patients from a public referral hospital in North-Central Ethiopia from July 2012 to January 2017. Log-rank and Wilcoxon tests, and an Accelerated Failure time (AFT) parametric statistical model were used.

Results: A mean of 22.13months (SD=13.16) and a median of 21.47months (IQR: 11.55-33.30) were found. The median age, baselineCD4 count, and BMI of patients at baseline were 30years (IQR: 25-38years), 247cells/µl (IQR: 120-375cells/µl) and 19.75kg/m² (IQR: 17.5-22kg/m²); respectively. At baseline, the patients CD4 count values ranged from 11 to 1764 (mean=289, SD= 233). The study revealed that 192(29.68%) events occurred, while 455(70.32%) were censored. The Gamma model was selected as the best-fit parametric model for the data. The estimated shape parameter of the fitted Gamma model is 1.9983 with 95%CI (1.3426, 2.9741) indicating that significantly determined as the survival time increases over time.

Conclusion: The fitted AFT parametric model (Gamma distribution), the factors such as: higher BMI, higher current CD4 count, being married, starting on AZT-3TC-EFV regimen class significantly increased the survival time of HIV-positive patients attending ART treatment, while; baseline CD4 count and ambulatory functional status reduces the survival time of HIV patients. Moreover, using ART treatment significantly improves the survival time of patients.

Keywords: AFT, gamma model, parametric, survival time

AFT, accelerated failure time; AIC, akaike information criteria; AIDS, acquired immune deficiency syndrome; ART, antiretroviral therapy; BIC, bayesian information criteria; BMI, body mass index; DBRH, debre berhan referral hospital; HAART, highly active anti-retroviral therapy; HIV, human immune deficiency virus; IQR, inter quartile range; SD, standard deviation.

Globally, the rapid distribution and high prevalence rate of HIV/AIDS among the working group of the population is a serious issue. HIV infection prevalence was highly observed in developing countries. Previously, due to the rapid increase of HIV infection in the 1980s and the 1990s, people with HIV have increasingly become infected. The epidemic has spread rapidly in towns and slowly in rural areas.¹

HIV epidemic in Ethiopia is becoming more concentrated in urban areas and along with the major transport corridors.² Currently, AIDS becomes the most leading cause of morbidity and mortality among adults in Ethiopia.³

According to the 2016-EDHS (2016-Ethiopian Demographic and Health Survey), HIV prevalence among men and women aged 15-49 is 0.9% [95%CI: 0.7%-1.1%],⁴ is lower than 2011 EDHS in which it was 1.9% for women and 1.0% for men with an overall prevalence rate of 1.5% [95%CI: 1.3%-1.7%],⁵ while it was 1.4% in 2005.⁶ The 2005 and 2011 EDHS reports show that the prevalence of the disease was significantly unchanged. The overall HIV prevalence rate in Ethiopia revealed inconsistent trends, with the majority of areas showing decreasing trends.⁷

Reports of 2005,⁶ 2011⁵ and 2016⁴ EDHS reports indicated that the prevalence rate of the disease varied from region to region, through socio-demographic factors, among urban and rural areas, and in the economic status of people’s. To safeguard the epidemic of the disease Ethiopia launched free ART treatment in 2005.⁸ Due to this, the country has observed remarkable progress over the past two decades in reducing HIV prevalence rate by 2.4%, from 3.3% in 2000 to 0.9% in 2017.⁹

For decades, scientists and humanitarian aid groups are working on pharmaceutical medicine for HIV/AIDS, but still not possible. The possible way, which is better than none- is treating patients through a clinical treatment called Highly Active Anti-Retroviral Therapy (HAART). Since HAART prolongs the lifetime of HIV/AIDS patients.¹⁰ Knowledge of HIV status helps HIV-negative individuals make specific decisions to reduce risk and increase safer sex practices to be free of the disease. For those who are infected with HIV, knowledge of their status allows them to take actions to protect their sexual partners, to seek treatment, and to plan for the future. Several cohort studies and clinical trials have shown that the CD4 count is the strongest predictor of subsequent disease progression and survival.¹¹

Even if there are barriers to HIV testing in Africa, including Ethiopia, from volunteer HIV testing through consultation to thinking as HIV is not a risky infection, from fear of being HIV-positive to fear of HIV stigma if HIV-positive, from no available cure to fear of no privacy and no confidentiality and no permission found from partner, especially for women;¹² the timely access of care and early diagnosis improves the clinical course of the disease and it reduces the transmission rates. The presence of a weak care system in low- and middle-income countries makes it difficult for the accessibility of HIV testing.

This study aimed to investigate the potential predictors of survival time in a cohort of HIV-infected patients on ART using the parametric survival statistical model perspective.

Study design, subjects and period

The study considers retrospective cohort study design on HIV-infected patients. The data used for this study were obtained from Debre Berhan Referral Hospital in the North Shoa Administrative Zone of the Amhara regional state of Ethiopia. The hospital serves as a referral center for different health centers in the surrounding districts. It has a unit for ART and gives Voluntary Counseling and Testing (VCT) services in a clinic for both referred and self-referred patients. Data were extracted from the patients’ chart, which has been adopted by the Federal Ministry of Health (FMoH) of the country, to be uniformly used by ART clinics to early identify and document laboratory, clinical and epidemiological variables. HIV-infected patients whose ages above sixteen years were included in the study. The study period was between the year 2012 to 2017 and the patients followed until one of the outcomes of event or censored occurred.

Inclusion and exclusion criteria

All HIV/AIDS-patients aged 16 years and above were included. Those patients who started ART before July 2012 or after July 2015 were excluded.

Data

Six hundred forty seven (647) patients, who fulfilled the inclusion criteria, were included in the study.

Data collection

Data were extracted from the patients’ medical charts by the ART clinic experts. The author had no accessed the information that could identify individual participants during or after data collection.

Variables

The dependent variable was survival time measured in months. The independent variables considered in this study were baseline age (in years), BaselineCD4 count, Sex (Male, Female), Marital status (Married, Single, Widowed/Divorced), Body Mass Index (BMI)(kg/m²), Functional status (Working, Ambulatory, Bedridden), current CD4 count, WHO clinical stages (Stage I, II, III, IV) and ART Regimen class (AZT-3TC-EFV, AZT-3TC-NVP, TDF-3TC-NVP, d4t-3TC-NVP, TDF-3TC-EFV).

Statistical data analysis

Statistical modeling of censored followed-up data in determining the potential predictor variables in the prognosis of survival time was commonly focused on employing nonparametric methods of Kaplan-Meier plots and log-rank test,¹³ semi-parametric method of Cox-Proportional Hazard (Cox-PH) model,¹⁴ and fully parametric method of Accelerated Failure Time (AFT) model.¹⁵ In the AFT parametric model, the survival time is modeled as a function of predictors or risk factors.¹⁶ This parametric model provides a more flexible result since the hazard rate is not constant with time.¹⁷ When comparing the Cox-PH model and parametric model results, parametric models lead to more efficient estimates.^18,19 The comparison is also done by several studies and the results showed that the parametric model fits the data.^20–22 The estimated coefficients in parametric models are interpreted as accelerating survival times.²³

In this study nonparametric method of Kaplan-Meier (KM) plots and log-rank tests,¹³ and the fully parametric method of Accelerated Failure Time (AFT)¹⁵ have been employed. Variable selection was undertaken based on the recommendations of Hosmer-Lemeshow (H-L).^24,25 Survival analysis is used to describe the analysis of data in the form of time from a well-defined time origin until the occurrence of some particular event or end-point.  If individuals did not provide complete information for a variety of reasons; subjects may not have been experienced the event of interest, in this case, the survival data can be censored. The existence of variables that change over time is a distinguishing feature in survival analysis.¹⁵

Model comparison and diagnostics

There are different methods of model selection, but for this study, we have used AIC and BIC criteria to compare various candidate models, and the model with the smallest AIC and BIC is selected as the better fit.²⁶ After the model is selected, the adequacy of the model was assessed using the Cox-Snell residuals for the parametric model.

Operational definition

Survival-time: the number of times from the date of enrollment to ART until one of the events of ”death”, ”lost-to-follow-up”, ”dropped-out”, ”stopped”, ”transferred out to other health care centers” occurred.

Statistical software

The data were coded and stored in SPSS version 20 and exported to SAS version 9.4 for analysis.

Exploring the data

The study included 647 individual patients in the treatment of HIV/AIDS from the year 2012 to 2017 for a total of 54 months in the hospital, with a mean of 22.13(SD=13.16) and a median of 21.47months (IQR: 11.55- 33.30). The median age, baseline CD4 count and BMI of patients at baseline were 30years (IQR: 25- 38years), 247cells/µl (IQR: 120-375cells/µl), and 19.75kg/m² (IQR: 17.5-22kg/m²); respectively. At baseline, the patients’ CD4 count values ranging from 11 to 1764 (mean=289, SD=233).

Of the total 647 patients, 423(65.38%) were females, while 224(34.62%) were males. The life-table estimates in the first six months indicates, about 11.34% of patients were failed, and the 54^th month is the riskiest, in which 29.68% were failed.

Out of the total subjects, 269(41.58%) were single, 217(33.54%) were married, and 161(24.88%) were had divorced/widowed marital status. Regarding functional status, 303(46.83%) were able to do their day-to-day activities (working), 142(21.95%) were ambulatory, and 202(31.22%) were bedridden. From the total 647 HIV patients, 224(34.63%) had started ART treatment with an initial WHO-clinical stage of stage I, and 126(19.47%), 176(27.20%), and 121(18.70%) were stage II, III and IV; respectively. At baseline, the ART regimen class of patients were AZT-3TC-EFV, 165(25.50%); AZT-3TC- NVP, 137(21.18%); TDF-3TC-NVP, 88(13.60%); d4t-3TC-NVP, 114(17.62%) and TDF-3TC-EFV, 143(22.10%) (Table 1).

Baseline socio-demographic, clinical and laboratory information vs. censoring status

Of the total 647 patients, 192(29.68%) were event occurred, and 455(70.32%) were censored. About 30.36% male HIV patients were event occurred, while about 69.64% were censored. And, on about 29.31% female HIV patients event were occurred, while; on 70.69% individuals censored were observed. There was no much difference between males and females in the occurrence of events. Regarding marital status, 36.80%, 33.18%, and 13.04%, married, single, and widowed/divorced patients events were occurred; respectively. In about half of the bedridden patients event were occurred (Table 1).

Univariable analysis

To determine whether there is a significant difference among different categorical covariates, we have employed the Log-rank and Wilcoxon tests of equality. The tests showed a significant difference is observed in survival situations among the categorical variables of marital status, functional status, WHO-clinical stages and regimen class subgroups (p-value=0.0001).

There were no significant difference in the time-to-event between sex (p-value=0.4587), WHO clinical stage (p-value=0.1301) and BMI (p-value=0.2310) for log-rank test of equality across strata for the predictors (Table 2). For other categorical predictors, the log-rank tests of equality across strata have the p-value of less than the cut-point of 0.25 or 25% significance level. 

Multivariable model

In identifying the better-fit parametric model to the data, we have been considered six parametric survival models: exponential, Weibull, normal, log-logistic, Gamma, and log-normal models. Comparison of parametric models was applied to find the most preferred model to represent the data parametrically. After a comparison of the models using AIC, BIC, and maximum likelihood ratio test values, the Gamma model was selected as the better fit with the smallest AIC and BIC value of 831.067, and the highest maximum likelihood ratio test value of -421.195 (Table 3). 

To infer the Gamma model parameters, we have specified dist=gamma in SAS proc lifereg.²⁷ The main limitation of Proc Lifereg in SAS is it fits for the generalized gamma model rather than the standard gamma model. There is a need for conversion of the generalized gamma model to the standard gamma model. We have applied the grid search technique to get the standard gamma distribution by using the outputs from the generalized gamma distribution (δ=σ). It was done for δ=0.3392 fitting a gamma distribution to the generalized gamma distribution with the dist=gammanoshape1shape1= 0.3392 no scale=0.3392 in SAS with log-likelihood Value -2419.191; for σ=1.9983 fitting a gamma distribution with the dist=gammanoshape1shape1=1.9983noscalescale=1.9983 with log-likelihood value -945.068. Then, the model with the largest log-likelihood value was selected as the best fit of the standard gamma model. The model with a scale and shape parameter of 1.9983 was selected as the better fit model for the standard gamma distribution. Therefore; the estimated gamma model scale/shape parameter is 1.9983 with 95%CI (1.3426, 2.9741), which is significantly indicating the survival time increase over time.  The plot of the hazard function looks like U-shaped because of δ=1.9983>1. The standard Gamma distribution parametric model parameter values are shown under (Table 4).

Based on the fitted AFT parametric model (Gamma distribution), the factors such as: higher BMI, higher current CD4 count, being married, starting on AZT-3TC-EFV regimen class significantly increased the survival time of HIV-positive patients attending ART treatment, while; baseline CD4 count and ambulatory functional status reduces the survival time of HIV patients.

Previous studies on HIV/TB co-infected patients in Nigeria²⁸ and HIV/AIDS patients’ survival in Ethiopia²⁹ results revealed that Weibull AFT model is the most appropriate. However, for this study the best fit for the data is Gamma parametric model. The estimated shape parameter of the fitted Gamma model is 1.9983 with 95%CI (1.3426, 2.9741) indicating that significantly determined as the survival time increases over time. Studies indicated that if the acceleration factor is greater than one, the survival rate has been increased compared with the baseline.³⁰ The plot of the hazard function looks like U-shaped, since (δ=1.9983)>1.

According to the multivariable Gamma parametric model, the covariates including BMI, baseline CD4 count, current CD4 count, functional status, marital status, and ART regimen class were significant at a 5% level of significance. In the multivariable Gamma regression model, the average survival time for females is exp(-0.1888)=0.83 [0.2234,3.0688], 17% lower than that of males. Females have a 17% decreased in survival time to males. In the present study, survival time for females was slightly worse than males after adjusting other covariates constants.  With A one-year increase in the age of the patient, the survival time of patients increased by exp(0.0570)=1.06 units or by about 6%. A unit increase in BMI of a patient would increase by 35% (e^0.2966=1.35) [95%CI: 1.1607, 1.5592]. It coincides with,³¹ in which patients with higher weight have less probability of dying than lower weight patients. Accordingly, a unit increase in the current CD4 count patient would improve the estimated survival- time by 1% (e^0.0055=1.01) [95%CI: 1.0019, 1.0092]. This coincides with,^31–35 in which it revealed that patients with higher CD4 counts have less probability of dying than patients with lower CD4 counts.

Moreover, married HIV-infected patients had higher survival time on ART than widowed or divorced patients, and ambulatory functional status HIV-infected patients increased the survival time by 0.3% than bedridden patients. Treatment of ART is effective enough in slowing down the progression of HIV-infection and decreasing the mortality rate of patients, and it is highly recommended to HIV patients start antiretroviral treatment early to track the progression of the disease.³³

The main aim of this study was to determine the potential predictors of survival time of HIV-infected patients on ART using a parametric survival model. The data best-fit with the Gamma parametric model. The estimated shape parameter of the fitted Gamma model is 1.9983 with 95%CI (1.3426, 2.9741) indicating that significantly determined as the survival time increases over time. The gamma model result shows that predictors including BMI, current CD4 count, ambulatory functional status, being married and began on ART regimen class of AZT-3TC-EFV were significant determinant factors of patients’ survival time at 5% significance level. Besides, using ART treatment significantly improves the survival of patients.

This study has a limitation of using some variables only. It is better for future researchers on HIV/AIDS patients on ART treatment will undergo their study by including more socio-demographic and economic variables.

None.

None.

The author declares that there was no conflict of interest.

1. Kloos H, Mariam D. HIV/AIDS in Ethiopia: An Overview. Northeast African Studies. 2000;7(1):13–40.
2. Federal HIV/AIDS Prevention and Control Office. HIV/AIDS in Ethiopia: An epidemiological synthesis. 2014.
3. Infection prevention guidelines for Healthcare Facilities in Ethiopia. Disease Prevention and Control Department AIDS, Addis Ababa, Ethiopia. 2004.
4. Federal democratic republic of Ethiopia. Ethiopia Demographic and Health Survey 2016: HIV Report. Ethiopia. 2016.
5. Demographic and health surveys. HIV/AIDS in Ethiopia: Data from the 2011 Ethiopia Demographic and Health Survey. DHS. 2012.
6. Central Statistical Agency and ORC Macro. Ethiopia Demographic and Health Survey 2005. Addis Ababa, Ethiopia. 2005.
7. Kibret GD, Ferede A, Leshargie CT, et al. Trends and spatial distributions of HIV prevalence in Ethiopia. Infect Dis Poverty. 2019;8:90.
8. Assefa Y, Jerene D, Lulseged S,  et al. Rapid scale-up of antiretroviral treatment in Ethiopia: successes and system-wide effects. PLoS medicine. 2009;6(4):e1000056.
9. Federal HIV/AIDS Prevention and Control Office. HIV Prevention in Ethiopia National Road Map 2018 – 2020 Final report. 2018.
10. The Republic of Uganda. National Antiretroviral Treatment and Care Guidelines for Adults, Adolescents, and Children, 2nd Edition. Ministry of health. 2008.
11. Mellors JW, Munoz A, Giorgi JV. Plasma viral load and CD4+ lymphocytes as prognostic markers of HIV-1 infection. Ann. Intern. Med. 1997;126(12):946–954.
12. Kwapong GD, Boateng D, Agyei-Baffour P,  et al. Health service barriers to HIV testing and counseling among pregnant women attending Antenatal Clinic; a cross-sectional study. BMC Health Serv Res. 2014;14:267.
13. Kaplan EL, Meier P. Nonparametric Estimation from Incomplete Observations. Journal of the American Statistical Association. 1958;53:457–481.
14. Cox DR. Regression models and life-tables. J R Stat Soc Ser B Stat Methodology. 1972;34:187–220.
15. Collet D. Modelling survival data in medical research, 2nd Edition. Chapman and Hall/CRC: London; 2003.
16. Kleinbaum DG, M Klein. Survival analysis: a self-learning text. Springer. 2005.
17. Stanley C, Molyneux E, Mukaka M. Comparison of performance of exponential, Cox proportional hazards, weibull and frailty survival models for analysis of small sample size data. Journal of Medical Statistics and Informatics. 2016;4(1):2.
18. Efron B. The efficiency of Cox’s likelihood function for censored data. Journal of the American statistical Association. 1977;72(359):557–565.
19. Oakes D. The asymptotic information in censored survival data. Biometrika. 1977;64(3):441–448.
20. Pourhoseingholi MA, Hajizadeh E, Dehkordi BM, et al. Comparing Cox regression and parametric models for survival of patients with gastric carcinoma. Asian Pacific Journal of Cancer Prevention. 2007;8(3):412.
21. Yazdani J, et al. Applying survival analysis to estimate survival time in gastric cancer patients. Journal of Mazandaran University of Medical Sciences. 2011;21(85): 28–36.
22. Nardi A, Schemper M. Comparing Cox and parametric models in clinical studies. Statistics in medicine. 2003;22(23):3597–3610.
23. Kleinbaum DG, Klein M. Survival analysis: a self-learning text. Springer Science and Business Media. 2005.
24. Hosmer DW, Lemeshow S, May S. Applied survival analysis, 2nd Edition. John Wiley and Sons, Inc: New York; 2008.
25. Hosmer DW, Lemeshow S. Applied survival analysis. Regression modeling of time-to-event data. New York; John Wiley and Sons; 1999.
26. Akaike H. A new look at the statistical model identification. IEEE Trans Autom Control. 1974;19(6):716–723.
27. SAS. SAS/STAT 9.2 User’s Guide. SAS Institute Inc. 2008.
28. Ogungbola OO, Akomolafe AA, Musa AZ. Accelerated Failure Time Model with Application to Data on Tuberculosis/Hiv Co-Infected Patients in Nigeria. American J Epidemiol Public Health. 2018;2(1):021–026.
29. Seyoum D, Degryse JM, Kifle YG, et al. Risk Factors for Mortality among Adult HIV/AIDS Patients Following Antiretroviral Therapy in Southwestern Ethiopia:  An Assessment through Survival Models. International journal of environmental research and public health. 2017;14(3):296.
30. Ghorbani N, Yazdani-Charati J, Anvari K, et al. Application of Weibull Accelerated Failure Time Model on the Disease-free Survival Rate of Breast Cancer. Iran J Health Sci. 2016;4(2):11–18.
31. Refera H. Survival and risk factors of HIV/TB Co-infected Patients under Antiretroviral Therapy in Ambo Hospital, Ethiopia. 2012.
32. Shewayiref Geremew, Dejen Tesfaw, Tibebu Getiye. Application of Longitudinal Measured CD4+ Count on HIV-Positive Patients Following Active Antiretroviral Therapy: A Case of Debre Berhan Referral Hospital. Biomedical Statistics and Informatics. 2018;3(2):34–42.
33. Gebremichael SG. AIDS-duration predictors of HIV/AIDS patients on antiretroviral therapy at Debre Berhan referral hospital, north-central Ethiopia. MOJ Public Health.  2020;9(4):99–105.
34. Bismark Sarfo, Naa Ashiley Vanderpuye, Abigail Addison, et al. HIV Case Management Support Service Is Associated with Improved CD4 Counts of Patients Receiving Care at the Antiretroviral Clinic of Pantang Hospital, Ghana. AIDS Research and Treatment. 2017;7.
35. Bhagat VK, Vinoth GCD, Kumar G, et al. Antiretroviral therapy adherence and its determinants among patients attending ART centre, Bhopal. Int J Community Med Public Health. 2018; 5:4566–4572.