Study Area and Period
Guraghe zone is one of the administrative zones in southern nations and nationalist region. It has 13 districts and two town administrations. Wolkite town is the capital of Guraghe zone which is found 425 km and 158 km from Hawassa and Addis Ababa respectively on the way to Jimma. There are five hospitals with four governmental and one private (Non-governmental) hospital. There were about 3032 adult ART clients on HAART in the zone [16].
Study design
A longitudinal cross sectional study design using retrospective review of charts (of adult ART clients on HAART) was conducted. Data were collected on randomly selected records of adult HIV positive clients on ART (age >18 years), from randomly selected Hospitals in Guraghe zone, southern Ethiopia.
Eligibility Criteria
Records of adult HIV positive patients on ART (age >18 years at enrollment) from the selected Hospitals, who have at least two consecutive BMI records during the follow up period were included in the study. While, records with incomplete data on outcome measurement; nutritional status indicators (weight, height) at different time periods of repeated measurements were excluded from the study. Cases which transferred to other facilities (transfer outs), in which the full follow up information was not available were excluded. As BMI is not appropriate indicator for pregnant women, those records of pregnant women without MUAC record were also not considered.
Sample Size Determination
The sample size for the first specific objective was determined based on single population proportion formula using magnitude of under nutrition among adult ART clients (P) and margin of error, 5% at 95% confidence interval. Hence, by using the prevalence estimates of under nutrition 10.5% (BMI < 18.5 Kg/m2) [17] the sample size became 144.
Even if mixed models have become the most popular method for analyzing repeated measures and longitudinal data; validated power and sample size methods exist only for a limited class of mixed models [19]. The appropriate sample size for the second specific objective was calculated by using significance level of 5%, at 95% Confidence level, with time interval of half a year, power of 80 %. By taking the larger sample size calculated using OR = 2.47, percent of unexposed with outcome (17.08 %), Type I error =5% [20], the sample size became 236. While, adding 5% loss of data and design effect of 2, the final sample size became 519.
Sampling Procedures
Out of the four public hospitals in Guraghe zone, two hospitals were randomly selected and included in the study. Out of the four hospitals, two hospitals were randomly selected with the fact that they are almost the similar regarding ART care (similar protocol) and patient characteristics. Then using the average five years’ adult ART case load of the two selected hospitals, the sample size was allocated proportionally. The records were selected using simple random sampling technique, by computer random number generator (using Open Epi software). The corresponding unique ART number (medical record number) of randomly selected clients’ record was used to retrieve the clients’ medical card from ART register.
Data Collection Methods and Procedure
Structured and cross checked data abstraction format was used to collect data from the client’s record at different periods of retrospective follow-up. The check list included information on relevant socio-demographic characteristics, weight, height, and treatment related issues. The data were collected by trained health professionals with basic skills and experience in ART documentation (intake form, follow up form and other medical records). Considering the total adult ART case load, the six Bsc nurses and public health officers with their supervisors were assigned to each selected hospital and collected the data. The primary outcome of this study is client’s repeated BMI score at different periods of follow up.
Study Variables
Dependent Variable of the study was Episodes of malnutrition among adults (Repeated measures) while the independent Variables included age, sex, employment status, functional status, nutritional therapy, ART adherence, baseline CD4, WHO stage, Cotrimoxazole prophylaxis, INH prophylaxis, Co morbidities, ART regimen, Presence of care giver, Opportunistic diseases, presence of eating problems and Gastro intestinal symptoms
Operational Definitions
According World Health Organization (WHO), malnutrition in this study was defined as low body mass index below 18.5 kg/m2 (< 18.5 kg/m2) which refers to under nutrition while those with BMI above 18.5 kg/m2 was considered as normal. Thus those with BMI < 16 (as severe malnutrition), 16 to 16.9 kg/m2 (moderate malnutrition) while those with BMI between 17 to 8.5 kg/m2 are defined as mild malnutrition or thinness [4].
Eating problems: for this study patients were considered as having eating problems, if they developed at least one of eating or swallowing difficulties due to oral hairy Leukoplakia, oral candidiasis, esophageal candidiasis or if they had loss of appetite at the time of the most current study period.
Data quality control
Two days training was given to the data collectors and supervisors before the actual data collection. Principal investigators and supervisors monitored and check the daily progresses. The information collected were cross checked with different sources (intake forms, ART register and ART chronic follow up form). Supervisors cross checked sample of daily collected data for correctness against the medical record or the ART register. Then reasonable feedback were given for the data collectors. The data were entered in to Epi data software and was restricted by legal values and other parameters to minimize errors. The data were entered by two independent data entry clerks and then cross checked for possible errors of data entry. Additionally, to keep the data quality, data from the medical records and the register were also crosschecked.
Data processing and analysis
The raw data were entered in to Epi-Data software version 3.1 and exported to SPSS version 25 for analysis. Data were presented in frequency, percentages, tables, and graphs. Nutritional status indicator for adults (BMI) was calculated by using compute command as weight divided by height in metre squared. Then the nutritional status was categorized in to under nutrition (below 18.5 kg/m2) and Normal (above or equal to 18.5 kg/m2) [4]. To assess the longitudinal episodes of malnutrition and its correlates, linear mixed model with random effect was done for repeated measure. In the meantime, the correlation between measurements of BMI was assessed. Thus, BMI score was considered as linear variable for linear mixed model. Spaghetti plot (with times of follow up on the X axis and the BMI score on the Y axis) was used to see the linear or curved pattern of the BMI scores. Multivariable random effects linear mixed models was fitted to estimate differences in BMI cell with respect to different factors (predictors). Maximum likelihood (ML) was used to estimate the parameter, as it considers both fixed and random factors. Under random effect model, parameter estimate with 95% Confidence Interval (CI) was calculated wit t and p value. Akaikes information criteria (AIC) was used to assess model fitness; with smaller value considered as complex in estimating parameter estimates. P-value of less than 0.05 used as cut off point to declare statistical significance.
Ethical consideration
Formal ethical clearance was obtained from the University’s Institutional Ethical Review committee and letter of cooperation was taken from the university to the zonal health office and to the respective hospitals sequentially. Before the data collection, informed consent was obtained (after full explanations of the study procedure) from the respective hospital managers. Then during the actual data collection all hard copy and softcopy data were under full protection in the hands of the investigators. The collected data were not used for other purposes than the study’s primary objectives. Client’s confidentiality was also kept.