Socio-demographic and Clinical Variables associated with the variation of CD4 Cell Count and Body Mass Index (BMI) for HIV Positive adults receiving HAART, a joint longitudinal data analysis

Objective The rate of prevalence of HIV/AIDS among adults has been increasing in Sub-Sahara African countries over the last decade. Currently, an estimated number of 722, 248 people are living with HIV, 23, 000 people are newly infected with HIV and 11,000 people are died because of AIDS related illness. The purpose of this study was to identify the most significant variables associated with the variation of CD4 cell count and body mass index (BMI) of HIV positive adults who initiated HAART at Felege Hiwot Teaching and Specialized Hospital, North-West Ethiopia. Methods To analyze the long-term CD4 cells count and body mass index of HIV infected adults, a prospective follow-up study of 792 HIV-infected patients was performed. A joint model was employed to identify the variables associated with the variation of CD4 cell count and body mass index of adults receiving HAART. A random of 792 samples was taken among charts in the hospital.

Socio-demographic, individual and Clinical variables had significant effect on CD4 cell count and BMI in HAART medication program. Time to follow ups in the HAART program had also direct and significant effect on the variables of interest.

Background
Currently, about 37.9 million people are living with human immunodeficiency virus (HIV) in the world and Sub-Saharan Africa carries the highest burden accounted for 71% of the global total [1]. In Ethiopia, about 722 248 people are living with HIV and the prevalence of HIV among adults is 1%. Hence, 23, 000 people are newly infected with HIV and 11,000 people are died because of AIDS related illness in [2]. Among all people living with HIV, 81% of them were on treatment and 73% of them are virally suppressed, 79 % knew their status, 65% are on treatment [3].
Of all adults aged 15 years and over living with HIV, 66% are on treatment [2]. Ninety-two percent of pregnant women in the country are living with HIV accessed antiretroviral medicine to prevent transmission of the virus to their baby. Women are highly affected by HIV as compared to men [4]. Hence, among 650000 adults living with HIV, 410 000(63.08 percent) are women [5]. New HIV infections among young women aged 15-24 years are more than double as compared to young men [6]. On the other hand, HIV treatment was slightly lower among women than men, with 55% of adult women living with HIV on treatment, compared to 66% of adult men [7].
Amhara region, one of the eleven regions in the country, had a prevalence rate of 1.6%, which is a serious case as compared to the national one. Hence, the region needs special attention to decrease the prevalence of HIV/AIDS and to reduce HIV related problems like low count of CD4 cell and being under weight of patients' body mass index (BMI) who are receiving HAART. Both medication and non-medication factors play a significant contribution for the variation of CD4 cell count and BMI of HIV patients. Both CD4 cell count and BMI are two of the strongest predictor of progression of HIV/AIDS and death [8].
Prevalence and factors associated with the variation of BMI in HIV-infected patients are reported and declared that the variation of BMI have a major psychosocial impact on such patients. A mixture of drug and non-drug-related factors are associated with such variation. Previous researches have shown that the variation of BMI is common among HIV-infected patients who receive treatment with antiretroviral drugs [8].
On the other hand, the amount of CD4 cell count provides a picture of immune system health with higher CD4 cell counts typically signifying healthier immune systems [9].
Naturally, children are born with high CD4 cell counts, which decrease slowly through adolescence and then plateau [2]. Demographic and genetic factors, exposure to additional infectious disease and behavioural factors have been associated with variations of CD4 cell counts in HIV-Positive adults [3]. HIV-positive adults with additional infectious disease have low CD4 cell count than those HIV-positive adults without other infectious diseases [8].
HIV adults who had good medication adherence are more probable to have more CD4 cell count [10]. Nevertheless, it is likely that communities in which the average income is high are those communities living in the urban areas [11]. Previous researches indicate that there are economic benefits associated with maintaining high CD4 cell counts for HIV positive people [12]. The adherence level of patients are highly associated with level of income such that patients with high level of income had good competence level of food and medication adherence [13].
Therefore, socio-demographic disadvantageous HIV positive adults are associated with lower CD4 cell count in the era of HAART [14]. The social factors are related to disclosure level of the disease and communication with people living together. The relationship between marital status and HIV infection is complex. In the previous research, it is indicated that patients aged 40 years and above presented lower rates of CD4 cell count change over the period of the cohort [15,16].
Successful strategies to get better retention in HIV care require an understanding of retention and adherence behaviour and the complex relationship between biological, psychological, behavioural, social and health system [17,18].
In Ethiopia, where the burden of HIV is the greatest, researches on factors affected variation of CD4 cell count of HIV-infected cohorts has extensively been done. However, detailed researches about socio-demographic, individual characteristics and clinical variables associated with the variation of CD4 cell count and BMI of HIV positive adults under HAART using advanced methods like joint models are limited [19]. Researchers conducted investigations separately for each of these outcome variables rather than joint effects. Furthermore, those employing methods described in current investigation are not properly described previously for the two responses. The knowledge and understanding of joint predictors of CD4 and BMI is important in a situation that large number of patients enrolled in HAART and further helps to reduce dropouts from the treatment. Joint models are more flexible in assessing the joint predictors of the variables of interests. Therefore, the objective of current investigation was to assess the joint predictors of two longitudinal response variables; CD4 cell count and BMI of patients who are receiving HAART at Felege Hiwot Teaching and Specialized hospital. This model had an advantage of gaining the efficiency of the parameter estimates. Hence, developing such model has methodological and practical contribution for health staff. The joint model of two longitudinal response variables can be used as reference for future investigation which implies for its theoretical The samples were selected using stratified random sampling technique, considering their residence as strata.
Data collection procedures: Secondary data were entered and analyzed using SAS software version 9.2. For the sample to be included in the study, CD4 cell count measurement and BMI just before the initiation of HAART were considered as a covariate so that there could be at least two visit responses after the initiation of their treatment for data analysis.

Measures of variables under study
In order to measure medication adherence and dietary instruction, patients were oriented to take pills after meal and these could be done on the prescribed time given by the health staff. At beginning of the treatment, patients directed to visit the hospital in each month and they took pills counted by the health staff sufficient for a month only. Patients also directed to bring back bottles of pills with unused pills. This was done for six months and then they directed to visit the hospital every quarter (three months) for the remaining study period. The reason for this was to follow up patients closely whether the new drug affects/ risks the health of patients at initial time. Hence, adherence was measured using pill count (adherence = pills used for given time/ total pills taken for the given interval *100%). In this regard, the health staff categorized a patient as medication adherent patient if he/she took at least 95% of the prescribed does, otherwise non-adherent patients. To assess whether, patients took meals on time prescribed by the health staff, self-reported data was considered. On top of this, health staff categorized patients as food adherent if he/she took at least 95% of the days in a month or in a quarter on the given time by the health staff. Similarly, to compute the BMI of patients, weight of patients were also recorded by the health staff and categorized underweight if BMI<18.5 kg/m 2 , normal if 18.5 < BMI< 24.9 and over weighted if BMI > 25 kg/m 2 . The patients' CD4 cells count was also measured at each visiting time using laboratory instruments (flow cytometry using anti-CD4 antibodies labelled with fluorescent dye) and recorded in chart of each patient. Self-reported data were also employed to assess whether there is social support from the community around them and for the existence of depressions on patients under HAART.

Data Analysis
In model selection, all predictors were considered in the model, and products of each predictor were fitted one a time to assess the interaction effect of predictors. To assess the separate estimation of BMI, ordinal logistic regression model was employed and a generalized linear mixed regression model (Quasi-Poisson) analysis was conducted to assess separate parametric estimation for CD4 cell count. In current investigation, the covariance structured and magnitude related to residual errors were also investigated for model selection, considering the model with smallest individual variability was the best and selected for data analysis. To assess alternative models in model selection, Deviance and Pearson chi-square were used to investigate model selection with the assumption that the model with smallest deviance is the best model. Finally, to assess the correlation between two response variables, CD4 cell count and BMI, the joint generalized mixed effect model was employed. In this model, the association between the two responses was specified through the random effect structure and then combining them by imposing joint multivariate distribution on the random intercept.

Results
Among the total patients registered at the Teaching and Specialized Hospital in Bahir Dar, 792 patients were randomly selected using patients' cards. The summary statistics of the baseline socio-demographic, individual characteristics and clinical variables of patients included in the study are indicated in Table1. Table1 shows that out of the sample of 792 patients: 40.9% were rural residents; 50.6% were females; 56.3% were living with their partner; 33.6 % disclosed their disease to family members; and 49.2% were owners of cell phones; 25.5% had good adherence to HAART; 11.5% had high incomes; and 20.6% had no education. Among the participants, on the average, 44% of them had under-weight baseline BMI.
In the analysis, patients who disclosed the disease reported that they got better social support from communities around them. To investigate this, HIV/AIDS stigma scale was used by the health staff at each visiting time. Among those patients disclosed the disease (266 patients), more than half of them (165 or 62% got social support. Similarly, depression of participants was also invented using Beck's depression inventory scale at each visit and 180 (22.7%) were depressed.
To assess the trend of dropouts, a logistic regression was conducted and the result revealed that dropouts were independent of results obtained from previous visit (chisquare = 0.324, p-value = 0.724). Hence, dropout did not have any reason given for progression rate of their previous visits which indicates that the trend of dropout was Missing Completely at Random (MCAR). Missingness were handled using multiple imputation technique.
The exploratory data analysis in current investigation indicates that as visiting time of a patient increased, the log of expected CD4 cell count also increased. However, the increasing rate of CD4 cell count was different for male and female patients, for urban and rural patients, for patients with different level of education, for patients who disclosed and not disclosed the disease for family members living together, for patients with different level of food and medication adherence and the like. On the other hand, as age of patients increased, the level of CD4 cell count decreased and the decreasing rate was also different for different categories of socio-demographic, individual, clinical and economic factors. Similar to CD4 cell count, the exploratory data analysis for BMI of patients was also investigated. The result shows that as visiting time of patients increased in taking treatment, the average BMI of patients also increased and the increasing rate was different for different groups. The SAS procedure that uses generalized mixed effect models allow the joint distribution to be constructed for the random effects from the two separate models.
The formulation of a joint model formed by striking a joint multivariate distribution helped in developing a joint multivariate distribution in the random effects of the two separate models. The joint modelling of the two responses was also conducted using the SAS PROC procedure and result of the two responses conducted simultaneously is indicated in Table4 and Figure2 show that, whenever, the number of follow-up visits of patients increased by one unit, the expected BMI for females was increased by 0.04 with p-value < 0.0001 and the log of expected CD4 cell count was increased by 0.05 with p-value < 0.01.
In Table4, as age of a patient increased by  Retention and close flow ups in medication care had also direct and significant effect on BMI which means patients which closely follow their prescribed medication given by the health staff had good adherence competence and this further leads to great BMI. This result agreed with previous researches [22,23]. Retention in medication care and adherence competence had direct and significant effect on the variable of interest (CD4 cell count change) which has similar argument with previous researches [21,22]. Retention in HIV medication care is a crucial activity for achieving optimal CD4 cell count and not to be under weight for patients following their treatment [24].
The economic factors such as patients with cell phone and those who had high income associated with high retention in the medication care. Hence, patients with high income may use different alternatives to get pills and he/she also uses proper food adherence schedules for the treatment to be effective and this encourages the patient to attend the visits of health institution and to have high BMI [25]. Patients' cell phone can play significant role in taking pills on time and to remind the date that the patient should visit the hospital and this has indirect effect on the status of CD4 cell count change. Cell phone helped patients to be HAART adherent because of its alarm (memory aid) for reminding the time that pills are taken and this also helps to gain high BMI [26]. This finding is consistent with findings from another study [27] and suggests the need for making cell phones available to the needy HAART attendants. However, this finding is contradicted with another previously conducted research [11]. Patients with high income and with ownership of cell phones belongs to urban areas and such patients had better CD4 cell count and normal BMI [28]. Previous researches also indicated that HIV positive adults who lived in rural areas exposed for malnutrition as compared to urban residents and this leads to be under weight and low CD4 cell count. Access to food or level of income is to a large extent positively correlated with BMI of patients. Hence, HIV patient with high income can have good accesses for food adherence and this further leads for good medication adherence [29].
Clinical factors such as patients' baseline CD4 cells count significantly affected their retention of medication care. High number of baseline CD4 cell count encouraged the patient to be good medication adherent and this further leads not to be underweight in HAART treatment [27,[30][31][32]. This result indicates that clinical factors (baseline CD4 cell count and medication adherence) positively associated with the follow ups of their treatment and this further leads for high BMI. Hence, HIV positive adults with high baseline CD4 cell count had high number of CD4 cell count/mm3. This result is similar with previously conducted research [22] but contradicted with another previously conducted study [29].

Conclusion
The result in current investigation corroborates key factors like realistic assessment of patients' knowledge/level of education and understanding of the regimen, residence area of patients, age and sex of patients. In this study, two longitudinal response variables (CD4 cell count change and BMI) were analyzed separately and jointly. The current study indicated that the variable of interest (CD4 cell count change) increased over time.
However, its progress was different for different groups. Consequently, due attention should be given to address the specific needs of each group of patients. Health related education should be given to patients to be good adherent and this leads for high progress of CD4 cell count and BMI [33,34]. Moreover, interventions need to be designed to promote both food and medication adherence and patients should be advised to disclose the disease to individual living together. Such individuals can remind them to take medication and food adherence on time. Health related issues like socio-demographic, economic, individual characteristics and clinical factors, public awareness through advocacy and social mobilization should be included in the ART service. Why the interaction effects existed in current investigation can be considered as a gap for future research.
One limitation of current investigation was that self reported data by the patients made uncertainty, since patients may discard unused pills and considered them as used once.
There was no means of control for the trustful of such data. However, the key factors incorporated in this study helped to identified groups of patients at risk and this further helps for intervention in HAART program. Competing interests: As no individual or institution funded this research, there is no conflict of financial interest between authors or between authors and institutions.

Authors' contributions
The first author wrote the proposal, develop data collection format, supervise the data collection process, analyzed and interpreted the data. The second and third authors participated in design and data analysis and critically read the manuscript and gave constructive comments for the betterment of the manuscript applying their reach experience. All authors contributed on manuscript preparation and discussed on order of authors.
Availability of data and materials: We confirm that the data used for this study is available at corresponding authors and can be submitted upon request.

Consent for publication:
This manuscript has not been published elsewhere and is not under consideration by any other journal. All authors approved the final manuscript and agreed with its submission. We agreed about authorship and order of authors for this manuscript.
Author's information: AST is an associate professor of statistics department at Bahir Dar University, Ethiopia with seven publications previously. Currently, he had finalized his PhD work in UNISA entitled "Modelling binary, ordinal and count response data; application of adherence and CD4 cell count change data" with the close supervision of the second and third authors. PN is a senior associate professor of the University of South Africa with more than 15 publications. TZ is a senior and full professor of college of mathematics, Statistics and Computer science at university of Kwazulu Natal. He has more than 130 publications in reputable journals. All the three authors together had five publications using the same data with the current one and on the same study area. This will be the six th article for our PhD paper. The previous five articles are; Figure 1 The plot of interaction effects between cell phone ownership and the number of follow-ups The plot of interaction effects between sex of patients and visiting times Plot of interaction effects between sex and age of patients