Serious adverse drug reactions in sub‐Saharan Africa in the era of antiretroviral treatment: A systematic review

Abstract We aimed to summarize and describe the burden of serious adverse drug reactions (ADRs) in sub‐Saharan Africa (SSA) in the era of antiretroviral therapy. We searched Medline, CINAHL, Africa‐Wide Information, Scopus, and Web of Science, without language restriction up to March 2021. We hand‐searched reference lists, conference abstracts, and dissertation databases. We included studies reporting proportions of admissions attributed to ADRs, admissions prolonged by ADRs, or in‐hospital deaths attributed to ADRs. Two reviewers independently screened the studies, reviewed the study quality using a previously published tool, and extracted the data. We tested for heterogeneity using I2‐statistics and summarized the study results using medians and interquartile ranges. Subgroup analyses summarized the results by study quality, setting, methodology, and population. From 1005 unique references identified, we included 15 studies. Median study quality was 7/10; heterogeneity was very high. Median [IQR] proportion of admissions attributed to ADRs was 4.8% [1.5% to 7.0%] (14 studies) and 6.4% [4.0% to 8.4%] in nine active surveillance studies in adults. Two pediatric studies reported the proportion of admissions prolonged by ADRs (0.29% and 0.99%). Three studies reported the proportion of in‐hospital deaths attributed to ADRs (2.5%, 13%, and 16%). Antiretroviral and antituberculosis drugs were often implicated in serious ADRs. Evidence of the burden of serious ADRs in SSA is patchy and heterogeneous. A few high‐quality studies suggest that the burden is considerable, and that it reflects the regional impact of the HIV pandemic. Further characterization of this burden is required, ideally in studies of standardized methodology.

disease and infectious disease, particularly HIV. The World Health Organization (WHO) first introduced guidelines for scaling up antiretroviral therapy (ART) in resource-limited settings in 2002. 7 Large national ART programs in SSA could potentially contribute significantly to the burden of ADRs in this region.
Serious ADRs are those that result in death, are life-threatening, result in hospital admission or prolong an existing hospital admission, result in persistent or significant disability or incapacity, or result in a congenital anomaly or cancer. 8,9 This systematic review aims to summarize and describe data on the burden of serious ADRs in SSA in the era of ART. We specifically focus on ADRs that cause hospital admission, prolong an existing hospital admission, or cause in-hospital death, as these three categories of serious ADRs are the ones most frequently measured by surveys, are reasonably easy to verify, are not subjectively judged, and do not require longitudinal data.
The specific objectives of this systematic review are: 1.1 | Primary objective 1. To summarize the proportions of hospital admissions attributable to ADRs, hospital admissions prolonged by ADRs, and in-hospital deaths attributable to ADRs in SSA in the ART era. 1. To summarize the proportions of hospital admissions attributable to preventable ADRs, hospital admissions prolonged by preventable ADRs, and in-hospital deaths attributable to preventable ADRs in SSA in the ART era.

| Secondary objectives
2. To describe common clinical presentations of serious ADRs, and drugs commonly implicated in serious ADRs, in SSA in the ART era.
3. To explore the contribution of HIV and ART to the burden of serious ADRs in SSA in the ART era. 4. To explore methodological and quality issues in ADR surveys conducted in SSA in the ART era.

| Criteria for considering studies for this review
We reviewed observational studies from SSA, published since 2002 (the year of the first WHO ART guideline for resource-constrained settings 7 ), which reported any of the following proportions: • the proportion of hospital admissions attributable to ADRs, • the proportion of hospital admissions prolonged by ADRs, or • the proportion of in-hospital deaths attributable to ADRs.
Prospective or retrospective cohort studies, cross-sectional studies, as well as data collected in the baseline survey / control arm of trials were eligible for inclusion. Forty-eight countries, listed in the supplement, were defined as SSA countries in accordance with the World Bank's use of the term. We used original study authors' definitions of "ADR", "hospital", and "admission", but we specifically did not consider attendance at an emergency unit to be a hospital admission. We only included studies conducted in unselected hospital populations. No restrictions were applied in terms of publication language or type; we included studies available as abstracts only. No restrictions were applied in terms of study population age groups.

| Search methods for identification of studies
We searched five databases (Medline, CINAHL, Africa-Wide Information, Scopus, and Web of Science) for relevant journal articles.
The primary search strategy, developed with the help of a medical librarian, was based on a combination of free text and index term searches for searching Medline through EBSCOhost. Terms identifying SSA were derived from a recommendation by the University of North Carolina Libraries, 10 and terms to identify ADRs were derived from a recommendation by the Cochrane Collaboration 11 and from terms used in a previous systematic review. 12 In addition to an "SSA concept" and an "ADR concept", we included two more search concepts, relating to the "seriousness" and the "prevalence" of ADRs.
Search strategies for all databases are included in the supplement.

| Screening of title/abstracts
After removing duplicates, two reviewers (JPM, and either NJ or GT) independently screened all reports on title and abstract. Reports were excluded if both reviewers agreed to exclude it; reasons for exclusion at this stage were not documented. Where no Englishlanguage abstract was available, we used an online translator (Google Translate) to assess the potential relevance of the report.

| Obtaining full-text articles
We were able to obtain full-texts of all the reports not excluded on title and abstract screening through the University of Cape Town Libraries; we did not need to contact authors for full-texts. Where multiple reports were found to relate to one study, we combined the reports into one study at this point.

| Full-text screening
Full-text studies were reviewed for inclusion independently by two reviewers (JPM, and NJ, GT, or KC). Studies in languages other than English were translated through an online translator (Google Translate). Disagreement over inclusion was resolved through discussion between the two reviewers, and a third reviewer could arbitrate. A reason for exclusion of the study was documented.

| Quality assessment
Included studies were independently assessed for quality by two reviewers (JPM and GT) using a slightly modified version of a qualityassessment tool developed specifically for ADR surveys. 12 Our modification replaced the term 'severity' in the tool with the terms 'seriousness or severity'. We calculated a quality score for each study as the total number of 'yes' responses out of tool's ten questions.

| Data extraction
Two reviewers (JPM and GT) independently extracted data on study characteristics, the study setting, the study population, the study methodology, and study findings. (The data extraction form is included in the supplement). Disagreement was resolved through discussion between the two reviewers, and a third reviewer (KC) could arbitrate. We did not contact study authors with data extraction queries.

| Data synthesis
Studies were grouped according to the data they contained relevant to this systematic review's three co-primary objectives. Group 1 studies reported the proportion of hospital admissions attributed to ADRs, group 2 studies reported the proportion of hospital admissions prolonged by ADRs, and group 3 studies reported the proportion of inhospital deaths attributable to ADRs; a study could be included in more than one group. Since findings could be reported on the level of the patient or on the level of the admission (i.e., allowing for readmissions), we decided to use admission-level data if reported, and patient-level data if no admission-level data were reported.
In the primary analysis, we pooled data from all studies in each group, regardless of study quality, study setting, methodological considerations, or study populations. We tested for heterogeneity using I 2 statistics to decide whether to conduct meta-analysis: in the presence of heterogeneity, we would summarize the proportions mentioned above as medians and interquartile ranges; in the absence of heterogeneity, we would conduct random-effects metaanalysis, calculating the pooled estimate proportion after Freeman-Tukey double arcsine transformation to stabilize the variances.
We explored the proportions mentioned above by subgroups relating to study quality, study setting, methodological considerations, and study populations. Depending on heterogeneity within each subgroup (using I 2 statistics) we would proceed to synthesize the data as above.
A similar approach was followed for the secondary objectives, where we summarized serious preventable ADRs. For the remainder of the secondary and explorative objectives, we narratively summarized clinical presentations commonly reported, drugs commonly implicated, and the contribution of HIV, ART, and methodological and quality issues.

| RE SULTS
Our electronic search yielded 1183 references and our hand search yielded three more; after deduplication 1005 references remained.
We excluded 964 references as irrelevant on title/abstract screening and assessed 41 studies for eligibility on full text. We excluded 26, listed in the supplement with reasons. The most common reason for excluding studies on full text was that no disaggregated numerator was reported. These included studies that did not distinguish between serious and non-serious ADRs, as well as studies of "umbrella topics", such as drug-related harm, which did not report ADRs separately. Fifteen studies [36][37][38][39][40][41][42][43][44][45][46][47][48][49][50][51][52][53] were included in this systematic review ( Figure 1). Quality scores ranged from 1/10 (in one study) to 10/10 (in four studies). Median [IQR] quality score was 7 [3 to 10]. All studies clearly reported the study design (Table 1), and most also clearly described data collection methods. In general, studies reported using standard methods for assessing causality, preventability, and seriousness/ severity; however, details of the processes applying these methods (for example, the people performing the assessments, solving disagreements, etc.) were less well reported. The study characteristic reported least often was the process of establishing seriousness. Table 2 summarizes the characteristics of studies included in this review. We investigated heterogeneity among subgroups based on study quality, study setting, study methodology, and study population. Since there was very high heterogeneity in each subgroup, no meta-analysis by subgroup was performed. Heterogeneity and summary proportions, by subgroup, are presented in the supplement.

| Group 1 studies
One study, reported only in two conference abstracts, 44,45 was a low-quality report of a spontaneous ADR reporting system in the medical wards of a Nigerian hospital, where 30/2012 (1.5%) patients were reported to have had an ADR-related admission.
The remaining 13 studies were all conducted as active surveillance studies, eight of which were in adults, [37][38][39][41][42][43]47,49,50,52 four in children, 36,40,51,53 and one in a mixed adult and pediatric population. 48 As this latter study presented data for adults and children separately, we could extract separate adult and pediatric datasets from it. Key results from Group 1 studies, grouped by the ADR detection method (spontaneous reporting vs. active surveillance) and study population (adults vs. children) are presented in Table 3.

| Adult active surveillance studies
Nine active surveillance studies contributed adult data ( Table 3). All were available as English full-text reports, except one Malian study reported in both French-language article 39 and French-language thesis 38 with English abstracts. An Ethiopian study was reported in two complementary articles. 49,50 Median quality score was 7/10, ranging from 2/10 to 10/10. Studies were conducted in South Africa (two studies, 37,47 both high quality), Nigeria (two studies 43,52 ), and in Eritrea, 48 Ethiopia, 49,50 Malawi, 42 Mali, 38,39 and Uganda. 41 Study duration ranged from 1 month to 16 months, for a median of 6 months. Six studies were single-center studies at secondary-/ tertiary-level hospitals, [37][38][39]42,43,49,50,52 while three were conducted in multiple hospitals, 41,47,48 including primary-level hospitals in 41,48 Where reported, all studies were conducted in medical wa rds, [37][38][39]41,43,47,49,50,52 additionally including intensive care units in one. 47 Only one study reported universal sampling of all patients admitted to the study wards 47 ; non-random sampling of admitted patients was described in the remainder. In these studies, large numbers of potential participants were often excluded (35%, 41 41,42,[48][49][50] ) and the WHO-UMC method (in three studies 37,47,52 ) were mostly used. Six studies reported assessing the preventability of ADRs, using Schumock and Thornton criteria in three studies 37,41,47 and unreported or other methods in three studies. [48][49][50]52 Apart from summary demographic statistics, the population included in studies' denominator was generally poorly described.
Mean or median age ranged from 36 to 50 years (reported in six studies [37][38][39]41,47,49,50,52 ), and the proportion of females from 42% to 56% (reported in seven studies [37][38][39]41,43,47,49,50,52 ). Only two studies reported the proportion of patients included in the denominator that were exposed to drugs before their admission. 47,49,50 Only two studies described the most common reason for admission, being malaria 41 and cardiovascular disease 47 respectively. HIV prevalence among patients included in the denominator was only reported in the South African studies, being 32% in the earlier 37 and 29% in the later 47 study. The proportion of patients in the denominator who were taking ART increased in the interval between these two studies from 5.2% 37 to 14%, 47 and was 11% in the Ethiopian study. 49,50 Among the nine active surveillance studies in adult populations, the median [IQR] proportion of admissions attributed to ADRs was 6.4% [4.0% to 8.4%]. Although more studies assessed the preventability of ADRs, only one study, from South Africa, reported the proportion of admissions attributed to preventable ADRs, which was 3.7%. 47 Four studies reported the drugs or drug classes implicated in ADRs causing admission to hospital, 37,42,47,49,50 with striking similarities: antiretroviral agents, antituberculosis therapy, cardiovascular drugs, and hypoglycemic drugs predominated. Common clinical presentations of these ADRs, as reported in five studies, [37][38][39]42,47,49,50 mostly related to their hepatotoxic and nephrotoxic effects, and hypoglycemia. Bleeds from non-steroidal anti-inflammatories and antithrombotic agents were less frequently reported. 47,49,50 Four studies reported HIV prevalence among patients who were admitted to hospital for ADRs. This was 38% in both South African studies, 37,47 28% in the Ethiopian study, 49,50 and 3/3 in the Malawian study. 42

| Pediatric active surveillance studies
Five active surveillance studies contributed pediatric data (Table 3), including two from Nigeria, 36,40 two from South Africa, 51,53 and one from Eritrea. 48 All were available as English full-text articles, with median quality score 5/10, although two studies 40,53 scored 10/10. Study duration ranged from 1 month to 36 months. Three studies were single-center studies at secondary-/ tertiary-level hospitals 36,40,51 while two were conducted in multiple hospitals, 48,53 including primary-level hospitals in one. 48 One study 53 surveyed patients admitted to intensive care units in addition to those admitted to pediatric wards. Only one study reported universal sampling of all patients admitted to the study wards 36 ; non-random sampling of admitted patients was described in the remainder. In these studies, unknown numbers of potential participants were excluded for reasons including no consent, 48,51 missing records, 40 short duration Only one study reported the proportion of patients included in the denominator that were exposed to drugs before their admission. 53 Infectious diseases were frequently reported as the reason for admission. 36

| Group 2 studies
We found no study conducted in adult populations reporting the proportion of hospital admissions prolonged by ADRs. Two pediatric studies, one from Nigeria 36 and one from South Africa 53 reported this proportion (

| Group 3 studies
Three high-quality adult studies, two from South Africa 37,46 and one from Ethiopia, 49,50 reported the proportion of in-hospital deaths to which ADRs contributed ( The three studies respectively found 2/80 (2.5%), 37  Multivariable logistic regression in one study 46 identified HIVinfection with antiretroviral treatment, higher drug count, and higher comorbidity score as independent risk factors for ADRrelated death. Unadjusted bivariate analyses in the other 49,50 also showed associations between ADR-related death and exposure to antiretroviral treatment, higher drug count, and higher comorbidity score, as well as pre-existing liver disease, a history of prior ADR, low body-mass index, and exposure to antituberculosis drugs.
In addition to the three studies included in this group, fatal ADR outcomes were reported by nine other studies included in this review. 36,[38][39][40]42,44,45,48,[51][52][53] However, these nine studies did not report the proportion of ADR-related deaths against a denominator of all in-hospital deaths.
A Nigerian study 52   Renal impairment (24), hypoglycemia (22), DILI (20), hemorrhage (19), blood dyscrasias (14) Rifampicin (17), enalapril (13), insulin (14), tenofovir (14), warfarin (Stevens-Johnson syndrome) in the other, 40 and 1/40 in a South African study. 53 No fatal outcomes were reported among 61 serious and nonserious ADRs in the other South African pediatric study. 51 In the country-wide Eritrean survey 48 fatal ADRs were reported among 5,848 patients admitted. 48 The two most common ADRs resulting in death were anemia (attributed to various drugs, including zidovudine) and hepatotoxicity (mostly attributed to antituberculosis therapy). Drugs used in the management of HIV, TB, and opportunistic infections appear to have been implicated in 17 deaths.

| DISCUSS ION
A small number of relatively high-quality studies report that seri- Study quality, as measured by the tool we applied, was generally high. Nevertheless, to fully comprehend the context within which ADRs occur, and the risk for their occurrence, it is essential to understand the denominator study population, including their diseases and their drug exposures. 64 Many studies included in this review did not clearly describe the denominator study population, and had this been included as a factor in the quality assessment, we may have reported lower quality overall. A further quality concern among nearly all studies included in this review is the high proportion of potential participants excluded from the surveillance activity. Nonparticipation was often over 50% and as high as 96% in one study, which may have resulted in a biased sample. Again, although this is a serious quality concern, it is not reflected in the quality assessment tool we used in this review.
We found very high heterogeneity among the study results, which can be attributed to high variability in study designs as described in Table 2, and also to high variability in the study settings and populations, potentially including some unreported differences. It is well-established that the reported prevalence of medication-related hospital admissions depends on the setting,

TA B L E 4
Key results from studies reporting the proportion of admissions prolonged by ADRs (Group 2 studies)
Another previous systematic review, published in 2016, aimed to compare adult ADR burdens in high-income countries and LMICs. 16 The authors found the median proportion of admissions attributed to ADRs to be lower in LMICs (5.5%) than in high-income countries (6.3%). 16 However, this review included only three studies from SSA among the 13 LMIC studies. 16 Our comparable estimate (median 6.4% among nine adult active surveillance studies) is probably more representative, and read together with the results from other systematic reviews mentioned above 1-6 probably dispels the idea that the burden of ADR-related admissions in SSA is lower than in highincome countries.
The proportion of in-hospital deaths attributed to ADRs in Europe was the topic of a 2021 systematic review. Six studies contributed 657 drug-related deaths out of 7578 in-hospital deaths, with the meta-analytic estimated proportion being 7.3% (95% CI 4.1% to 12.5%). 65 This estimate appears to agree with an earlier populationbased study using linked databases in Sweden, in which 6.4% of inhospital deaths were attributed to ADRs. 66 We identified only three studies conducted in SSA to report this particular outcome, and these reported 2.5%, 13%, and 16% of deaths in adult medical wards were attributed to ADRs. The low number of SSA studies reporting this proportion as outcome precludes any meaningful interpretation of this proportion. However, it is notable that in two of the studies a large proportion of the deaths were associated with preventable ADRs. In addition, it is notable that most deaths were due to renal and liver injuries, with ART and antituberculosis therapy most often implicated. This stands in contrast with studies from Europe, which found hemorrhages to be most common fatal ADRs. 65,66 For several of our study objectives, we found limited data. Only two studies reported on admissions prolonged by ADRs, and there were minimal reports on admissions caused by or prolonged by preventable ADRs. These are knowledge gaps to be filled by future research.
Another exploratory objective yielding little data was describing the role played by HIV and ART in the serious ADR burden in SSA. Although rarely reported, it was consistently reported that HIV prevalence among patients with serious ADRs was higher than HIV prevalence among the denominator population. 37,46,47,49,50,53 This imbalance was most pronounced for the outcome of ADR-related deaths. However, this finding should be interpreted with care.
Where antiretroviral therapy was implicated in causing ADRs, these were often older antiretrovirals, including stavudine and efavirenz, and the ADRs were often unpredictable and unavoidable events.
ART programs in SSA are continuously improving and introducing newer drugs with fewer toxicities: stavudine has been phased out as first-line ART option in the period 2006 to 2011, 67 and efavirenz is currently being phased out in favor of dolutegravir. Tenofovir alafenamide has better renal safety than tenofovir disoproxil fumarate 68 and may in future replace it in ART programs.

| CON CLUS ION
We have shown that evidence of the burden of serious ADRs in SSA is patchy and highly heterogeneous. Nevertheless, a few high-quality studies suggest that the burden is considerable. A unique feature of the ADR burden in this region is the frequency at which people living with HIV appear affected, and the frequency with which ART and medicines used in the management of opportunistic infections are implicated in this burden. However, the risk of ADRs should be considered against the risks associated with non-treatment, 64 and there is no doubt that the benefit of these medicines massively outweigh their risk of harm. Further characterization of the serious ADR burden in SSA is required, particularly in pediatric and elderly populations, and in countries other than South Africa and Nigeria. This should ideally be performed by conducting studies of standardized methodology.
As a first step toward limiting the risk of harm caused by medicines, the focus should be on avoiding preventable ADRs, which are often caused by errors. Investing in support systems that catch errors before harm occurs is required, particularly in under-resourced and overburdened health care settings such as those in SSA.

CO N FLI C T O F I NTE R E S T
The authors have no conflicts of interest to declare.

AUTH O R CO NTR I B UTI O N S
JPM and KC conceived and designed the study. JPM conducted the searches. JPM, NJ, GT, and KC screened studies for inclusion. JPM, GT, and KC extracted the data and assessed the study quality. JPM synthesized the data, conducted the analyses, and wrote the initial draft of the manuscript, under the supervision of KC. All authors critically reviewed and revised the manuscript.