HIV treatment and monitoring patterns in routine practice: a multi-country retrospective chart review of patient care [version 2; peer review: 1 approved, 2 approved with reservations]

Background: A study of patient records in four HIV clinics in three sub-Saharan African countries examined routine clinical care patterns and variations. Methods: Clinic characteristics were described, and patient data extracted from a sample of medical records. Data on treatment, CD4 count and viral load (VL) were obtained for the last visit in the records, dates mainly between 2015 and 2017, patient demographic data were obtained from the first clinic visit. Results: Four clinics, two in Nigeria, one in Zambia and one in Uganda, all public facilities, using national HIV treatment guidelines were included. Numbers of patients and health professionals varied, with some variation in stated frequency of testing for CD4 count and VL. Clinical guidelines were available in each clinic, and most drugs were available free to patients. The proportion of patients with a CD4 count in the records varied from 84 to 100 percent, the latest median count varied from 269 to 593 between clinics. 35% had a record of a VL test, varying from 1% to 63% of patients. Lamivudine (3TC) was recorded for more than 90% of patients in each clinic, and although Open Peer Review


Introduction
In addition to choice of appropriate drugs, best practice management of patients with HIV requires monitoring response to treatment and disease progression. This includes tracking clinical immunological, and virological data on patients at diagnosis and on follow-up. There is a rich literature guiding HIV treatment, and guidelines are developed and updated regularly as new evidence comes to light [1][2][3][4][5] (see http://www.who.int/hiv/pub/guidelines/en/ for an historic list of guidelines produced by the World Health Organisation). The implementation of standardized protocols for treatment and investigations may vary in resource constrained countries, with differences in resources available for such services. For example, the current guidelines for antiretroviral therapy (ART) requires all patients on ART to have viral load test at six months and 12 months 5 , implementation might vary based on availability of resources to support viral load testing. While a number of reports of current treatment and testing practice are available from resource-limited settings, these are mainly in the context of patients enrolled in research centres or in research studies 6-8 . Such settings often have substantially more resources than routine clinical practices, and as such practice patterns may differ from routine practice, where issues such as clinic and patient resources or drug stock-outs can affect care. This study was designed to explore variation in the evaluation, treatment, care and follow-up among patients diagnosed with HIV in routine care settings in low-to middle-income countries.

Methods
Interest in participating in this study was sought among the Masters of Public Health Alumni of Peoples-uni 9 . A protocol was developed by those who responded (Supplementary File 1), and a retrospective cohort was created via standardized record review.
A data collection instrument was developed, based on previous research and publications and management guidelines [1][2][3][4][5] . A spreadsheet was created with coding instructions (Supplementary File 2). In addition, data on the characteristics of the setting for each facility were collected in early 2017 by each investigator in consultation with local clinic staff, and entered onto an online survey form (Supplementary File 3). This included country and city, hospital or other healthcare facility, number of patients seen, and what diagnostic, treatment, referral and follow-up facilities are available. Clinic and patient data were de-identified to maintain confidentiality. Four clinics from three countries in sub-Saharan Africa chose to participate in the study.
For clinic BA, a public hospital clinic in northern Nigeria, patients were selected in sequence of attendance as new patients, starting January 1, 2013 for whom records were also available over the next two years. Patient demographics were those obtained at the first visit, and CD4 count and ART treatments were recorded at each visit, with analysis relating to data recorded at the most recent visit.
For clinic EV, a public hospital clinic also in Nigeria, all existing patients were re-tested with an ELISA method in 2013/14, and patients were randomly selected for this study among those who tested positive at that time. CD4 count and treatments recorded were those at the latest visit, and patient demographics were those in the records from their first visit.
For clinic MW, a public community-based clinic in Zambia, patients were randomly selected from those present on the patient registry in 2015/16. At that time, patients were reviewed for the need to start on highly active antiretroviral therapy (HAART) based on CD4 count, and the ART regime recorded was that started as a result of that review at that time. Data on patient demographics were those present in the records, often this would be prior to the date of entry to this study.
For clinic AM, a public hospital based clinic in Uganda, the sampling frame was the register of all clients on active ART, and every third patient file was retrieved from files ordered according to clinic appointment date. The CD4 counts and ART regimes recorded were for the most recent measures in the records, and patient demographics were those in the records from their first visit.

Measurement of study factors
Data on individual patients were extracted from medical records, including age, gender, and baseline clinical data at diagnosis and follow-up. Clinical information included clinical, immunological and virological information on patients at diagnosis and on follow-up, prescription of ART and other drug choices, important co-infections and/or co-morbidities.

Measurement of outcome factors
Details of the tests ordered and their results, and treatments ordered were extracted from medical records. Individual patients were not contacted. Data were obtained from the records, and no attempt was made to validate the information. Missing data were recorded as missing, and not explored further.
A pilot study tested the feasibility of the data collection and the method of recording on to a spreadsheet. Data were collected by a research assistant in each setting, using the spreadsheet, from examination of individual records.

Ethics requirements
As a retrospective patient records study, consent was not requested from individual patients. Ethics approval was sought and obtained in each setting from the appropriate authority (see Ethical approval and consent section). Considerable care was taken not to reveal the identity of any individual and all data were deidentified. The spreadsheet for the recording of data only had an

Amendments from Version 1
The revised version has amended Table 2, added recent references, provided more information on the dates of the CD4 counts and viral load tests, and discussed the impact of different caseloads on outcome.

REVISED
identification number, and the key to the identity of each patient was kept separately to maintain confidentiality. To further this, the clinics names and exact locations have been removed.

Sample size considerations
Each centre was asked to obtain information from at least 100 patients.

Data analysis
Descriptive statistics were used to characterise the study population. Data distributions were assessed, checking for skewness and kurtosis. Data summary statistics were generated. Categorical variables were summarized using proportions while continuous variables were summarized using medians and interquartile ranges. Due to differences in data extraction across sites, no statistical analyses comparing across sites was conducted. All analyses were conducted in Stata 14.1 (StataCorp, College Station, X) and R version 3.3.0 (The R Foundation for Statistical Computing).

Results
Ten alumni expressed interest in participating, three centres were able to pilot the data collection instrument and four centres in three countries participated in the data collection for the study. Overall, data were abstracted for 600 patients. Table 1 shows the characteristics of the 4 clinics. The number of patients seen per month varied from 500 to 4200, and the number of doctors, nurses, and allied health professionals available at the clinics varied from 1 to 50, 10 to 65, and 6 to 45 respectively. There was some variation in the stated frequency of testing for CD4 count (either every three or six months) and for measurement of viral load. Each clinic had availability of clinical guidelines, and most drugs were available free of charge to patients. All clinics have access to ART and co-trimoxazole. Table 2 shows the patient characteristics and test results. Age, gender and body mass index (BMI) were similar between clinics, with overall median age of 31 years, 62% were female and median BMI was 20. The proportion of patients with a CD4 count in the records varied from 84 to 100 percent, and the latest median count varied from 269 to 593 between the clinics. 209 (35%) had a record of a viral load test with 81% of then having a viral load of less than 50 copies/µL. Only 23 (19%) of 119 patients had records of AIDS defining illnesses. Figure 1 shows the variation in the distribution of the CD4 counts in each of the four clinics. As reported in the records, 58% of the latest CD4 counts were from 2016 or 2017, 27% from 2015, and the remaining 15% earlier; 95% of the viral load tests were from 2016 or 2017. There was no information about the timing of the reports being given to the patients themselves. Table 3 shows the treatment regimes. Lamivudine (3TC) was recorded for more than 90% of patients in each clinic, but otherwise there was considerable variation between clinics in the choice of ART. Tenofovir disoproxil fumarate (TDF) was prescribed for 317 (53.9%) of studied patients. Efavirenz   (EFV) was prescribed for 326 (55%) of the patients, making combination of TDF/3TC/EFV the most common ART combination used by studied patients. In two clinics, 100% and 99% of patients were prescribed co-trimoxazole prophlaxis, compared with 7% and no patients in the two other clinics. There were 6 patients on lopinavir and 8 on atazanavir (second line treatment choices). Scheme used to code the data is available as Supplementary File 2

Discussion
Data were collected in a standardised way, with data on treatment regimes and CD4 counts relating to the latest information in the records (dates mainly between 2015 and 2017). While we initially planned to collect data on a cohort of patients enrolled two years before data collection in each clinic, this did not prove feasible and patients had presented to the clinics at variable times.
This study found modest documentation of clinical activities, with a wide variation in available clinic health work force, frequency of CD4 counts and levels and viral load assessment. The two clinics in Nigeria had markedly different caseloads. The centre with the high caseload is one of the oldest in Nigeria and provides incentives to this population of patients (free medical care to family members, distribution of food supplements and sometimes fares, etc.), Differences in case load may reflect population distribution and heterogeneity in HIV prevalence, or that patients may seek care far away from their base to avoid stigmatization. Variation in case load can lead to differences in efficiency and quality of care, and potentially could have implications for care outcomes. However determining its impact on outcome would a require a properly powered longitudinal study.
While most patients had a CD4 count in the records, there was a wide spread of latest CD4 counts, with both within-and betweencentre variation in the CD4 counts. Only one in three patients had a record of viral load at any time during the course of treatment, and most of these had an undetectable level (<50 copies/µL). Some of the absence of viral load might be poor record keeping, and the high proportion of undetectable results might reflect an absent rather than a low test result, and hence not an indicator of treatment success. Over the years the CD4 threshold for commencing ART had been lowered, as can be seen in Table 4, with the most current guideline requiring a "test and treat approach" ie eliminating CD4 threshold as a prerequisite for commencement of ART 5 . While this guideline is based on sound evidence, it adds an unanticipated number of potential eligible patients for ART care, with attendant strain on countries with an already fragile economy. The low frequency of viral load test results might also reflect resource limitations.
Most patients were on TDF/3TC/EFV drug regimen, which is consistent with current guideline advice for first line treatment. Only 2% of the patients were on second-line ART, varying from 9% to less than 1% between clinics. It is likely that most of the studied patients are still on a potent first line ART judging by the low second-line ART usage. Use of a regimen with a low pill burden would enhance therapy adherence and might lead to reduced need for switching patients to second line ART 10 . TDF/3TC/EFV drug regimens have a low pill burden, and may contribute to the low usage of second-line ART regimens.
The variation in the use of co-trimoxazole, from almost universal in two clinics, to negligible in the other two clinics, is an extreme example of the variation we identified. The reason for this variation despite the guideline recommendation is not clear, but it is possible that it is a result of local clinic policy settings, possibly related to concerns about increasing antibiotic resistance. Variation in the use of co-trimoxazole has been found in other resource-limited settings, while efforts are being made to improve the rates of use 11 . While each clinic reported having access to treatment guidelines, those current at the time of the data collection do not appear to have been universally followed. The guidelines also do change regularly, as shown in Table 4. For example in relation to the frequency of CD4 count monitoring and the CD4 count threshold at which treatment should be started have changed since the time relating to the study data -new guidelines recommend starting treatment regardless of the CD4 count 5,12 . Some discrepancy between WHO guidelines and actual implementation in practice may arise from the time it takes to implement new guidelines, or due to lack of resources to immediately initiate all patients with HIV on ART. We see considerable between-clinic variation in a number of key management strategies reported by the clinics, from the recording of CD4 counts, median CD4 counts on treatment, and treatments used. This variation is consistent with a previous survey of stated management practices in 6 sub-Saharan countries 13,14 .
The management of patients in routine clinical care has been shown to differ from that seen in clinical trials 15 , as well as to lead to worse clinical outcomes, although it is beyond the scope of our study to explore outcomes. Findings of variation from standard management practice has previously been reported from routine care settings in other parts of sub-Saharan Africa including Ethiopia 16 , Uganda 17 and Tanzania 18 , mostly in single isolated centres. Here we present combined data on health system related measures across multiple sub-Saharan African HIV treatment sites.
In the absence of standardised record keeping systems, it is difficult to make clear comparisons of management and outcome in routine clinical care. Our findings suggest that there is room for further work to understand the reasons for these record gaps, and to standardise the record keeping in routine care of HIV positive patients 19-21 . The potential of electronic medical records to improve records could be explored 22 .
Our study has a number of limitations. First, this study was an analysis of data extracted from existing medical records, which are prone to error including missing data. We have found missing data where it should ordinarily not be missing. We cannot comment on other factors that may have contributed to missing data, such as whether tests were not done or if they were not recorded. Second some of the variation between centres may be due to differences in patient populations which we were unable to capture, even though we selected clinics involved in routine clinical care. Third, participation in the study was restricted to few countries, and to individual clinics, which may not be representative of the national picture in these countries. A prospective cohort study on representative samples would provide a more robust study design with more detailed quantitative data to delineate care dynamics and to provide longitudinal data to better understand how clinical practice in these settings is linked to patient outcomes.

Conclusions
We demonstrate a wide variability in compliance with HIV treatment guidelines in four routine care settings in sub-Saharan Africa, as well as gaps in the records available. The findings of this study may provide an explanation for heterogeneous HIV treatment outcomes across sub-Saharan Africa. In spite of the limitations, these data underscore the need for an in-depth study to address compliance with HIV treatment guidelines and best practice. While electronic medical record implementation might be a challenge for many HIV care points in sub-Saharan Africa, our findings emphasize the need for more robust interim paper-based medical record keeping.

Ethics approval and consent
Ethics approval was sought in each setting from the appropriate authority. For two of the centres, research ethics committees gave approval. In two of the centres, the ethics committees stated that they did not require formal approval from them, however approval to access records was obtained from the relevant District Health Office/r. Details for each clinic as follows: Clinic AM: Eastern Uganda AIDS Support Organization (TASO) contacted who recommended no need for an approval but rather write to the Amuria District Health Officer (DHO) for permission to have access to the hospital records, which was given.
Informed consent was not obtained from individual patients since this was a records study with appropriate institutional approval, strict confidentiality arrangements, and no patient contact, as stated in the manuscript.

Data availability
Dataset 1. De-identified data collected from clinical records used to create Table 2 and Table 3 and Figure 1. Scheme used to code the data is available as Supplementary File 2. 10.5256/ f1000research.15169.d206333 23 All data underlying the survey responses from clinic informants for Table 1 are available as part of the article.

Grant information
The author(s) declared that no grants were involved in supporting this work.

Supplementary File 1 -Study protocol
Click here to access the data.
Supplementary File 2 -Data coding scheme Click here to access the data.
Supplementary File 3 -Written from of online survey Click here to access the data.

If applicable, is the statistical analysis and its interpretation appropriate? Partly
Are all the source data underlying the results available to ensure full reproducibility? Yes Are the conclusions drawn adequately supported by the results? Partly Response: As indicated in the paper (Methods) 'Interest in participating in this study was sought among the Masters of Public Health Alumni of Peoples-uni' and (Results) 'Ten alumni expressed interest in participating, three centres were able to pilot the data collection instrument and four centres in three countries participated in the data collection for the study. ' We have added to the sentence in the Methods section as follows: "Interest in participating in this study was sought among the Masters of Public Health Alumni of Peoples-uni. There was no prior determination of the numbers of centres required or their geographical setting." We have added to the section on study limitations as follows: "Since this is a descriptive study and not hypothesis testing, it was not feasible to determine sample size in advance. A pragmatic approach was adopted, with each centre expected to obtain information from the first 100 patients presenting over a 6-month period, or from at least 100 patients." We have changed the first sentence in the Discussion to the following: "Data were collected according to the standardised data collection instrument (Supplementary File 2), with data on treatment regimes and CD4 counts relating to the latest information in the records (dates mainly between 2015 and 2017)." We believe that the background information that the Referee requires on the centres chosen is fully reported in Table 1 ' We have added the following in the study limitations section in response to the Referee's request: "A future study should perform an audit of the medical records to determine the reason for missing data and hence the potential effect on the study findings." In response to the Referee's request to identify the new information added by our study, we have added the following sentences to the Discussion section (paragraph 7) to follow our original sentence 'Here we present combined data on health system related measures across multiple sub-Saharan African HIV treatment sites.' the additional sentences read: "Hence, this study adds to the literature a current examination of the routine care provided to patients with HIV, rather than that in the context of a clinical trial. The study adds information from a number of centres in multiple countries, using a common protocol, of both the characteristics of the clinics and of the care given in these clinics." I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.

1) Is The study design appropriate and is the work technically sound?
The study design wand the work is technically, as the authors endeavoured to analyse multicountry settings 2. Are sufficient details of methods and analysis provided to allow replication by others?
The methods and analysis were sufficient, however in the case of Nigeria where there are 2 health facilities with different case loads per month; the authors could also enlighten us on the impact of having a high and low case load as presented.
3. If applicable, is the statistical analysis and its interpretation appropriate?
The statistical analysis was appropriate, although the authors should also enlighten us on whether the records indicate the dates when the CD4 count and viral load tests were taken and when they were reported back. I suppose this of great importance since, there are reports of patients who have their blood samples taken and never receive results.
Thank you for your generally positive comments. In the revised version submitted, we have added four recent references, and given further details on the dates of CD4 counts and viral load tests (although we have no data on the time that the results were reported to the patients). We have also discussed the reasons that the two Nigerian clinics might have different caseloads and the potential impact of this.

Mahidol Oxford Tropical Medicine Research Unit (MORU), Bangkok, Thailand
This article by Musa and colleagues uses a snapshot of patient records from four HIV clinics in Nigeria, Zambia and Uganda to describe the variability in a number of patient measures. Data are not reported longitudinally and all analysis is descriptive.
The most useful part of this report is to show that ordinary clinics in countries with some of the highest HIV burdens globally do not confirm with WHO guidelines. For instance, only commencing ART when CD4 cells <350/ul. Further, the high amount of missing data in viral loads suggest there are logistical or cost challenges to routine testing.
It is perhaps unsurprising that clinics in different countries vary in their patient characteristics and levels of reporting. The variability in CD4 counts is not particularly informative as it is not linked to length of time under ART. No research questions are addressed. Despite the descriptive nature of the study, the authors are to be commended on presenting hard to access data from routine health care settings in sub-Saharan Africa. I hope that they will be able to take longitudinal data in the future and better understand how clinical practice in these settings is linked to patient outcomes.

Is the work clearly and accurately presented and does it cite the current literature? Yes
Is the study design appropriate and is the work technically sound?