Use of hospital services by patients with chronic conditions in sub-Saharan Africa: a systematic review and meta-analysis

Abstract Objective To estimate the prevalence of individual chronic conditions and multimorbidity among adults admitted to hospital in countries in sub–Saharan Africa. Methods We systematically searched MEDLINE®, Embase®, Global Index Medicus, Global Health and SciELO for publications reporting on patient cohorts recruited between 1 January 2010 and 12 May 2023. We included articles reporting prevalence of pre-specified chronic diseases within unselected acute care services (emergency departments or medical inpatient settings). No language restrictions were applied. We generated prevalence estimates using random-effects meta-analysis alongside 95% confidence intervals, 95% prediction intervals and I2 statistics for heterogeneity. To explore associations with age, sex, country-level income status, geographical region and risk of bias, we conducted pre-specified meta-regression, sub-group and sensitivity analyses. Findings Of 6976 identified studies, 61 met the inclusion criteria, comprising data from 20 countries and 376 676 people. None directly reported multimorbidity, but instead reported prevalence for individual conditions. Among medical admissions, the highest prevalence was human immunodeficiency virus infection (36.4%; 95% CI: 31.3–41.8); hypertension (24.4%; 95% CI: 16.7–34.2); diabetes (11.9%; 95% CI: 9.9–14.3); heart failure (8.2%; 95% CI: 5.6–11.9); chronic kidney disease (7.7%; 95% CI: 3.9–14.7); and stroke (6.8%; 95% CI: 4.7–9.6). Conclusion Among patients seeking hospital care in sub-Saharan Africa, multimorbidity remains poorly described despite high burdens of individual chronic diseases. Prospective public health studies of multimorbidity burden are needed to generate integrated and context-specific health system interventions that act to maximize patient survival and well-being.


Introduction
As life expectancy increases in sub-Saharan Africa, so too does the number of people who live with chronic conditions.][3] Inequalities in access to health care for chronic conditions affect early detection and control, and therefore on healthy life expectancy.Where primary care provision is limited, the index presentation of chronic disease is commonly through hospital acute care services. 4,5Acute medical services in these contexts traditionally have a single disease focus and may overlook multimorbidity in vulnerable patients.
In sub-Saharan Africa, the burden from chronic diseases is projected to increase: an estimated 125 million adults will have hypertension by 2025; 6 and 26.9 million adults will have diabetes by 2030. 7Although dramatic reductions in the incidence and mortality of human immunodeficiency virus (HIV) have been observed in sub-Saharan Africa over the past 30 years, with increasing life expectancy, the high prevalence of HIV infection is presenting new challenges and demands within existing health-care systems. 8As such, integration of multimorbidity care into hospitals in sub-Saharan Africa will be of increasing importance over coming years.Cohort studies of adults in community settings have reported prevalence of multimorbidity of 69% (absolute numbers not available) in South Africa and 65% (252/389) in Burkina Faso. 2,9However, data on the prevalence of individual chronic diseases and multimorbidity in sub-Saharan African hospital settings are limited. 10o estimate prevalence of chronic disease within unselected cohorts of adult patients admitted to medical wards and emergency departments within sub-Saharan Africa, we conducted a systematic review of observational epidemiological studies.We focused on hospital rather than community presentations as populations in sub-Saharan Africa commonly have limited access to primary care.As such, hospital presentation represents an important node of intervention to control chronic disease and prevent development of secondary complications.Development of prevalence estimates within the region are important for policy-makers to prioritize and optimize service design and care delivery in sub-Saharan Africa. 9,11atients with chronic conditions in sub-Saharan African hospitals Stephen A Spencer et al.

of an unselected inpatient population in either emergency departments or medical wards). Data on outcome conditions are available (Box 2).
Exclusion criteria were paediatric populations; community or out-patient settings (not acute care); denominator not available for population of interest (e.g.selected disease-specific cohorts or patients recruited solely from renal or cardiology wards); mental health conditions; trauma or surgical conditions; maternal, obstetric or gynaecological conditions; behavioural risk factors (excluding alcohol and tobacco); conference abstracts.
We excluded paediatric populations as patterns and clustering of multimorbidity in children younger than five years has been reviewed elsewhere, and found to be different than adult populations. 16Similarly, multimorbidity in maternal care in sub-Saharan Africa has recently been examined, suggesting a specific analysis for non-pregnant adults would complement these efforts. 17e restricted studies to those published and conducted since 1 January 2010 to avoid the use of data before the accelerated roll-out of antiretroviral treatment (ART) in sub-Saharan Africa which has driven changes in disease patterns. 18he 2010 cut-off is aligned with reporting frames of the Joint United Nations Programme on HIV/AIDS (UNAIDS) and Global Burden of Disease (GBD) studies. 8,18,19We did not apply language restrictions to inclusion criteria.

Databases and search terms
We systematically searched MEDLINE®, Embase®, Global Index Medicus, Global Health and SciELO databases on 12 May 2023 for articles published since 1 January 2010.EndNote X9.3.3 software (Thomson Reuters, Eagan, United States of America) was used to export references, and to identify and remove duplicates.Box 1 shows key search terms; the full search strategy is available in the online data repository. 20

Selection process and data collection
Two authors independently assessed article titles, abstracts and full manuscripts to select studies meeting the eligibility criteria.Subsequently they piloted and refined the data collection tool 20 using the first five eligible studies.These two authors then independently and manually extracted data from each manuscript, and assessed for bias using the modified Newcastle-Ottawa Scale for non-randomized studies 21 (online repository). 20We categorized scores of ≤ 3 as very high risk of bias; 3-6 as high risk of bias; and scores of 7-9 as high quality. 22Discrepancies in selection and bias decisions were resolved through discussion and arbitration by a third reviewer.

Analysis
We captured extracted data using Microsoft Excel (Microsoft Corporation, Redmond, USA) and analysed using Stata 15 (StataCorp LLC, College Station, USA).We assessed publication bias by visual inspection of funnel plots of prevalence data when > 10 prevalence estimates were included, 23 and Egger's test. 24To visualize and assess individual disease prevalence alongside both 95% confidence and prediction levels, we gener-ated forest plots with meta-analyses. 25e chose random effects modelling a priori due to the expected high level of heterogeneity, 26 and we calculated pooled confidence with heterogeneity by the Hartung-Knapp-Sidik-Jonkman method. 27,28Data were logit transformed except when close to the extreme boundaries, where Freeman-Tukey doublearcsine transformation 29 was employed (the full STATA statistical analysis code is available in the online repository). 20eterogeneity was assessed by I 2 statistic and by 95% prediction intervals (95% PI) which estimate the range of values in which future similar studies would be expected to fall. 30Random effects models were used to calculate 95% PIs when ≥ 5 study estimates were included in the meta-analysis, due to the high degree of imprecision with very low numbers of estimates. 31,32e identified chronic conditions contributing to potential multimorbidity in adults in sub-Saharan Africa from the Global Burden of Disease 2019 databases for risks (risk factor conditions) and causes (diseases). 14The top 15 common causes resulting in death were included.In addition, HIV treatment failure and non-compliance were included a priori as significant drivers of HIV morbidity and mortality. 15

Primary outcomes
Prevalence of the pre-selected primary preventive conditions and secondary (end-organ) conditions

Primary preventive conditions:
HIV, hypertension, diabetes, obesity, alcohol use, smoking and dyslipidaemia

Secondary (end-organ) conditions:
Stroke, ischaemic heart disease (including acute coronary syndrome), heart failure, chronic liver disease, chronic kidney disease and chronic obstructive pulmonary disease (COPD) Secondary outcomes: (i) prevalence of multimorbidity in acutely unwell adult patients presenting to hospitals in sub-Saharan Africa; (ii) prevalence of decompensated chronic disease-associated admission; and (iii) prevalence of HIV treatment failure, HIV treatment non-compliance, undiagnosed HIV and HIV status awareness.Patients with chronic conditions in sub-Saharan African hospitals Stephen A Spencer et al.
We performed meta-regression analysis where > 10 prevalence estimates were present per condition. 33We included a priori within univariable meta-regression analyses: age (median or mean); sex; date of study.In view of continued expansion in ART availability across sub-Saharan Africa, 34 we also examined the temporal changes in the prevalence of conditions through meta regression.We also used meta regression to assess the association between studylevel HIV prevalence and country-level adult HIV prevalence (as given by GBD 2019 database for adults ≥ 20 years). 14e pre-planned to report all prevalence estimates stratified by hospital population (medical vs emergency department); country-level income status defined by the World Bank 2022 Fiscal Year; 35 geographical regions defined by the African Union (Central, East-ern, Southern, and Western Africa); 36 and Newcastle-Ottawa-Scale.We also planned in advance to report prevalence estimates among the high-and midquality graded studies through sensitivity analyses once studies with a very high risk of bias were removed.

Results
We identified 6976 manuscripts, of which 61 studies met the inclusion criteria (Fig. 1).These articles included 17 prospective cohort studies; 37-53 11 retrospective cohorts; [54][55][56][57][58][59][60][61][62][63][64] and 33 crosssectional studies.  The oled sample size was 376 676 participants, including 97 737 participants admitted to the emergency department, and 278 939 admitted to medical wards.We did not identify any studies that intentionally investigated prevalence of multimor-bidity as a primary objective.We have therefore structured the results section to explore the prevalence of the most commonly identified individual chronic diseases, followed by a section exploring available data on multimorbidity from secondary analyses of included studies.
Characteristics of studies that met the inclusion criteria are described in
Due to limited emergency department data (< 10 studies), only data from medical wards were included in the meta-regression and sub-group analyses.HIV infection prevalence among medical in-patients correlated with national HIV prevalence (odds ratio, OR: 1.33; 95% CI: 1.09-1.63;online repository). 20Higher HIV prevalence was noted in southern Africa (46.0%; 95% CI: 40.5-51.7),as compared to eastern Africa (22.5%; 95% CI: 19.8-25.4).There was no association between HIV prevalence and year of study (OR: 0.93; 95% CI: 0.84-1.04);or country-level income status, sex or average age (online repository). 20Prevalence of HIV was also unaffected by the removal of studies with a very high risk of bias (36.9%; 95% CI: 31.6-42.6).Patients with chronic conditions in sub-Saharan African hospitals Stephen A Spencer et al.

Reporting bias
We found evidence of publication bias from small studies reporting prevalence of hypertension (Egger's P -value: 0.04), but no evidence of publication bias for other conditions (online repository). 20o reduce very high risk of bias, we conducted sensitivity analyses, and eight studies with very high risk of bias were subsequently excluded from the metaanalyses, with no observed changes in synthesized prevalence estimates (online repository). 20

Discussion
Here we present synthesized prevalence data for multiple individual chronic diseases among hospitalized adults in sub-Saharan Africa.We found no studies that directly measured the prevalence of multimorbidity, although secondary analyses within these studies suggest Fig. 3 Our estimated HIV prevalence in sub-Saharan African hospitals is about eightfold higher than reported estimates at the community level in sub-Saharan Africa (36.4% versus 4.7%). 14Reassuringly, more than 90% of HIV-infected patients included in this review knew their diagnosis; however, treatment failure in about one third of patients indicates that viral control should be a keystone issue for future public health campaigns.Absence of temporal changes in our review may reflect regional and sub-national variability in the HIV epidemic and ART scale-up. 8,34his result contrasts data from Malawi which shows falling HIV admissions from 2012 to 2019. 56or hypertension, prevalence in sub-Saharan communities is estimated at 30.8%, 99 which is similar to hospital prevalence found in our study (24.4%).We found that admission with severe uncontrolled hypertension was higher than in high-income countries (5.2% versus 1.9%). 100With diabetes, the estimated hospital prevalence was 11.9% which is higher than community levels in sub-Saharan Africa (4.2%). 14Diabetic emergencies represented 40% of patients admitted with diabetes.In contrast, findings from the National Diabetes Inpatient Audit England 2019 found that diabetic emergencies were approximately one in 20 in diabetic inpatients. 101In sub-Saharan hospitals, the high burden of decompensated illness presentations indicate missed opportunities to better diagnose and control disease.
A similar pattern was seen in disease burden from end-organ complications, dominated by heart failure and stroke (8.2% and 6.8%, respectively), which are higher than estimates from outside sub-Saharan Africa (1-2% 102,103 and 3.7-4.4%). 104,105This reinforces ob-servations that in sub-Saharan Africa: (i) hypertension is the leading cause of heart failure and stroke; and (ii) 88% of the global hypertension mortality is found in low-and middle-income countries. 106Estimates from the Global Burden of Disease 2017 suggest that ischaemic heart disease is the most common cause of cardiovascular-related death in sub-Saharan Africa (5% of all deaths). 107][110][111] The strengths of our study include studies reporting data from unselected medical and emergency department populations, designed to reduce selection bias.There were no language restrictions in our search strategy, and we were able to include data from 20 sub-Saharan African countries, representative of nearly 400 000 patient admissions.We explored heterogeneity by calculating 95% prediction intervals to provide clinically relevant information on the degree of heterogeneiety. 30n addition, we restricted our pooled synthesis and used robust methods to explore potential explanations, including predetermined sensitivity, subgroup, and meta-regression analysis.
The heterogeneity observed in our analysis is a common limitation of systematic reviews of disease prevalence. 7,112The reasons for this include differences in population demographics; criteria and tools used to ascertain outcomes; and study quality.Given differences between countries in terms of access to health care, socioeconomic status, geography and ethnicity, heterogeneity both between and within countries in sub-Saharan Africa is expected.For example, HIV prevalence is likely to be higher in hospitalized patients compared to the general population for a given country, with multiple factors (e.g.success in meeting the UNAIDS 90-90-90 objectives) 113 influencing this relationship.
Another limitation was the nonuniform application of diagnostic criteria, and likely inconsistent access to laboratory assays, equipment and technical expertise.Quality issues relating to outcome ascertainment were identified in over half of all included studies.For instance, troponin, electrocardiogram and angiography were underutilized in the diagnosis of acute coronary syndrome, and spirometry for chronic obstructive pulmonary disease.This underutilization may have led to underreporting of these conditions.Although 13.1% of studies were at very high risk of bias, sensitivity analyses demonstrated consistent disease prevalence estimates.
A key finding from this systematic review is the lack of primary outcome data on multimorbidity in sub-Saharan hospitals.Synthesized community-level data from predominantly high-income settings have estimated multimorbidity prevalence is 33.1%, 114 with disease combinations reflecting the most prevalent individual long-term conditions within the population. 115Prospective cohort studies, designed explicitly to examine multimorbidity prevalence using standardized diagnostic tools and criteria, could support the development of health services more responsive to patient need.
We found high prevalence of single chronic diseases in hospital settings.From the limited data on multimorbidity identified within the secondary analyses of included studies, it is probable that there is a high burden of missed multimorbidity in sub-Saharan Africa.When examining the secondary outcome data from the studies included in this analysis, it was revealed that there may be a significant burden of multimorbidity in this particular context.For instance, one study primarily focused on investigating the prevalence of hypertension among medical inpatients, but it also discovered that out of the 59 patients with hypertension, 33 of them were also diagnosed with diabetes. 66We also found high prevalence of acute decompensated presentations.We observed increased chronic disease prevalence within hospitals compared to community settings.Hospitalized patients in sub-Saharan Africa are therefore likely to have increased preventable disability and early mortality compared to high-income settings.
Our review suggests important clinical and policy implications.Similarly, the need for context-appropriate diagnostics was underscored by a 2023 World Health Assembly resolution. 116ull World Health Organ 2023;101:558-570G| doi: http://dx.doi.org/10.2471/BLT.22.289597   Patients with chronic conditions in sub-Saharan African hospitals Stephen A Spencer et al.
Inconsistent use of diagnostic tools and criteria has also been described within the Lancet commission on diagnostics, showing limited or no access for 47% of the global population. 117Implementation of standardized chronic disease programmes which focus on community care (e.g. the WHO package of noncommunicable disease interventions) 118 could be strengthened by explicit linkages to secondary clinical pathways.
Successful implementation of such linkages will require broad health systems approaches including: healthcare worker training; development of financial models that promote reliable access to diagnostics and essential medicines; integration with existing health information systems; 119 robust governance structures; and strengthened local leadership. 116The need to shift away from disease-to patient-centred approaches is a consistent theme highlighted in recent Lancet commis-sions. 120,121Improved health literacy is likely to empower patients and their caregivers in managing their health and chronic conditions, and navigating care pathways.Policies which link primary and secondary care for chronic disease management could facilitate more accessible and cost-effective models of care delivery, from both provider and patient perspectives.■
Patients with chronic conditions in sub-Saharan African hospitals Stephen A Spencer et al.Studies recruited participants with acute medical illnesses in emergency departments and triaged for medical care.We have therefore categorised these studies as prevalence estimates among patients from medical wards.
The estimated glomerular filtration rate is calculated using the Chronic Kidney Disease Epidemiology Collaboration equation.Note: The average age and the percentage of females reflect values from the unselected hospital population (either emergency department or medical wards).Where studies have not reported figures representative of the unselected population, we have marked this as NR (not reported or not representative of the unselected emergency department or medical ward population. (. ..continued)

Fig. 1 .
Fig. 1.Flowchart showing the selection of studies included in the systematic review on patients with chronic conditions in sub-Saharan Africa

Conditions, context, population criteria and search strategy used for the systematic review on patients with chronic conditions in sub-Saharan Africa Criteria Conditions
Box 1. : Chronic diseases or risk factors that are likely to contribute to multimorbidity.Context: Acute admission to adult medical wards or emergency departments in hospitals in sub-Saharan Africa.Population: Adults of both sexes that meet the 'context' criteria (above).

Table 2 . Prevalence data of chronic health conditions and risk factors in patients admitted to medical wards or emergency departments, sub-Saharan Africa Condition, by patient population No. of patients (no. of studies) Prevalence, % (range) 95% CI 95% PI Between group heterogeneity, P Patients in medical wards
Bull World Health Organ 2023;101:558-570G| doi: http://dx.doi.org/10.2471/BLT.22.289597

. Prevalence of secondary end-organ conditions in patients admitted to medical wards or emergency departments, sub-Saharan Africa Patients in medical wards
CI: confidence interval; PI: prediction interval.Note: Values are also presented in Table2.Some conditions have no 95% PI because prediction intervals were calculated when ≥ 5 prevalence estimates were included in the meta-analysis Patients with chronic conditions in sub-Saharan African hospitalsStephen A Spencer et al.

Table 1 . Included studies in the systematic review on patients with chronic conditions, sub-Saharan Africa Study Country Data collection year Site Study design Outcome and outcome assessment Sample size Age of patients, years No. of female patients (%)
Patients with chronic conditions in sub-Saharan African hospitalsStephen A Spencer et al.

year Site Study design Outcome and outcome assessment Sample size Age of patients, years No. of female patients (%)
Patients with chronic conditions in sub-Saharan African hospitalsStephen A Spencer et al.

Study Country Data collection year Site Study design Outcome and outcome assessment Sample size Age of patients, years No. of female patients (%)
Patients with chronic conditions in sub-Saharan African hospitalsStephen A Spencer et al.Patients with chronic conditions in sub-Saharan African hospitals Stephen A Spencer et al.

Country Data collection year Site Study design Outcome and outcome assessment Sample size Age of patients, years No. of female patients (%)
CT: computed tomography; DBP: diastolic blood pressure; EEG: electroencephalogram; ELISA: enzyme linked immunosorbent assay; HIV: human immunodeficiency virus; IQR: interquartile range; mmHg: millimetres of mercury; MRI: magnetic resonance imaging; NR: not reported or not representative of unselected hospital population; SBP: systolic blood pressure; SD standard deviation.of Diet in Renal Disease equation, as modified by the criteria set by the Kidney Disease Improving Global Outcomes, is used to calculate the estimated glomerular filtration rate.
a The Modification b