Heart Failure Care and Post-Hospital Outcomes in Northwestern Tanzania

Study setting and population

We conducted a retrospective analysis of HF patients admitted to the adult medical ward at Bugando Medical Centre (BMC) between January 1 and December 31, 2023. BMC is an academic quaternary care center in the city of Mwanza. It serves as the referral center for eight regions across the Lake Zone of Tanzania, with a population of approximately 22 million people. Patients were eligible for inclusion in this study if they were diagnosed with HF at any point during their admission to the adult medical ward; there were no exclusion criteria. A diagnosis was made by the admitting or treating clinician; per standard clinical practice in Tanzania, this diagnosis was based upon the Framingham criteria (11).

The study was reviewed by the institutional review board at the Catholic University of Allied Health Sciences Bugando and the Weill Cornell School of Medicine. Since this is a retrospective review of de-identified existing clinical data, informed consent was not required from participants.

Study procedures

At the time of HF diagnosis during admission to the adult medical ward (patients aged >13), a physician and member of the cardiology care team captured the patients’ medical record numbers and up to three separate telephone numbers per standard protocol at BMC. Telephone numbers were for the patient and their designated contact(s).

During a one-month period (September 2025), data were abstracted from the electronic medical record by a cardiologist and member of the study team (Cody Cichowitz) using a data abstraction tool programmed into REDCap. Data were abstracted from the time of the index admission in 2023 and all subsequent care encounters occurring up until August 31, 2025.

Data abstraction

Data abstracted from the index admission included age, sex, health insurance status, the results of diagnostic testing, vital signs, and discharge medications. All notes from the index hospitalization were reviewed together with the echo report. Participants were categorized as having HF with reduced ejection fraction (HFrEF: left ventricular ejection fraction <40%), HF with mildly reduced ejection fraction (HFmrEF: left ventricular ejection fraction between 41–49%), and HF with preserved ejection fraction (HFpEF: left ventricular ejection fraction >50%). Available laboratory data included the hemoglobin at the time of admission and all creatinine and potassium levels measured during hospitalization. Comorbid chronic conditions captured included hypertension, diabetes, atrial fibrillation or flutter, and HIV. The diagnosis of comorbid conditions was made by the treating physician at the index hospitalization and abstracted from the electronic medical record using notes and ICD codes. Data abstracted from post-discharge encounters included vital signs, basic laboratory data, HF prescriptions, any subsequent readmissions, and in-hospital deaths.

Heart failure care metrics

We characterized the proportion of patients who received quality care as stipulated by current HF guidelines (7, 8, 9). We then enumerated the proportion of patients who received appropriate diagnostic testing with an echo and measurement of serum creatinine and potassium; were diagnosed with HFrEF and prescribed GDMT prior to discharge; attended post-discharge follow-up within one month of discharge; were retained in outpatient care three months after discharge; received GDMT for the first three months after discharge (HFrEF patients only); and had repeat measurement of creatinine and potassium within the first three months following discharge and GDMT initiation.

Within each class of medication used in GDMT, appropriate use was defined as the medication being prescribed and filled by the pharmacy or there being a contraindication to using that class of medication as described in the PEN-plus guidelines (7, 8, 9). For example, a patient with HFrEF was considered to have appropriate use of beta-blockers if they were either (a) prescribed or (b) not prescribed, and the heart rate was less than 55 beats per minute. Similarly, ACEi and ARBs are contraindicated in the PEN-plus guidelines if serum creatinine is greater than 2.3 mg/dL or the systolic blood pressure is less than 100 mmHg, while MRAs are contraindicated if the serum creatinine is greater than 2.3 mg/dL or the serum potassium is greater than 5.5 mEq/L. Additionally, sodium-glucose co-transporter 2 inhibitors were not widely available in Tanzania in 2023, nor covered by insurance. Thus, GDMT in this study was defined to be combination therapy with beta-blockers, ACEi or ARBs, and MRAs. Assessment of GDMT would ideally include evaluation of titration, but we observed that very few participants were prescribed GDMT consistently. Therefore, we did not include medication titration in this analysis.

Post-discharge events

Following medical record abstraction, all patients in the registry were contacted by a physician and member of the cardiology care team at BMC (Mariam Goba) using the phone numbers provided during their index admission, along with any additional numbers documented in their medical record. To maximize response rates, each participant was called on three separate days at different times of day. If a participant or their family member was reachable, vital status was recorded, along with any known hospital readmissions, ongoing care engagement for HF, and medication use.

For the subset of patients who could not be reached by phone, two additional steps were taken: For patients with health insurance, we contacted the health insurance company to ascertain vital status. For patients who were not reachable and did not have health insurance, two members of the study team independently reviewed all available electronic medical records—including severity of heart failure, comorbid conditions, vital signs, and medication prescription data—and categorized vital status as ‘likely dead’ or ‘unknown’; any disagreements were adjudicated by a third reviewer. Because phone numbers in Tanzania only remain active if they are used every three months, and each number is connected to mobile banking, it is unlikely that patients would lose access to all mobile numbers if they were still alive.

Data analysis

Standard descriptive statistics were used to characterize patients admitted with HF at BMC during 2023. We described inpatient and post-hospital mortality and evaluated factors associated with death during the index hospital HF hospitalization or after discharge. Univariate and multivariable logistic regressions were performed to assess for factors associated with inpatient mortality. Cox proportional hazards regression analyses were performed to assess for factors associated with post-hospital mortality. Exposures of interest were defined a priori and included age, sex, insurance status, timing of HF diagnosis (i.e., established or new diagnosis), HFrEF, anemia, chronic kidney disease, hypertension, diabetes, and atrial fibrillation or flutter. The proportional hazards assumption was assessed using the Grambsch–Therneau test based on scaled Schoenfeld residuals, evaluated both globally and for each individual covariate, and supplemented by visual inspection of scaled Schoenfeld residual plots against time and log-minus-log survival curves. For models exhibiting non-proportionality, sensitivity analyses incorporating time-varying covariates were performed to assess the temporal stability of the hazard ratios. Complete data were available for each of these exposures except for ejection fraction (11.9% unavailable), kidney function (7.3% unavailable), and hemoglobin (5.6% unavailable); data were unavailable because these diagnostic tests were not completed. In the primary analyses, missing data were assumed to be normal. We performed two secondary analyses: one restricted to participants with complete data and the other wherein missing data were assumed to be abnormal.

In the primary analysis of post-hospital mortality, participants who were not reachable over the phone were censored at their last known care encounter. We conducted two pre-specified sensitivity analyses. In the first, patients were considered dead if (a) death was confirmed during data collection or (b) they were out of care, not reachable, and their vital status was determined to be ‘likely dead’ during vital status adjudication by the study team. In the second sensitivity analysis, we assumed that all participants who were out of care and not reachable were dead.

We next assessed whether each discrete step in HF care was associated with post-hospital mortality using univariate and multivariate Cox proportional hazards regression analyses, adjusting for the same variables included in the base model evaluating for factors associated with post-hospital outcomes. In the regressions that assessed for an association between attending the first post-discharge appointment and retention in care at three months and survival, we restricted the analyses to only include people who survived at least one month and three months after hospital discharge, respectively. GDMT utilization was further characterized at the time of hospital discharge, any subsequent readmissions, and from all outpatient clinic visits. All analyses were completed in STATA 19.

Study population

Between January 1 and December 31, 2023, a total of 177 patients were admitted to BMC with a primary diagnosis of HF (Table 1). The median age was 61.0 (IQR: 45.9, 72.8) years. Women comprised half of the cohort (50.8%). Among the patients, 59.9% were without health insurance/self-paying, and 79.7% were diagnosed with HF for the first time during index hospitalization. Approximately half of patients (48.6%) had HFrEF, 12.4% had HFmrEF, and 27.1% had HFpEF; 11.9% of patients admitted with HF never received an echo. Hypertension (46.9%) and chronic kidney disease (CKD) at Stage 3 or above (65.0%) were the most common comorbidities, followed by diabetes (18.1%) and atrial fibrillation or flutter (17.5%).

Heart failure outcomes

Of the 177 patients who were admitted with HF, 23 (12.9%) died during the index admission, and 154 (87.1%) survived to discharge (Table 1). Among the 154 patients who survived to be discharged, vital status was ascertained for 110 patients by phone call or hospital/insurance records review. For the remaining 44 participants, the study team adjudicated outcomes as ‘likely dead’ for 37 patients and ‘unknown’ for 7 patients. There were high rates of early mortality (Figure 1); 24 months after discharge, 63.6% of patients were dead (98 out of 154; 61 confirmed deaths and 37 likely deaths).

Figure 1

The presence of renal dysfunction on admission was the only exposure independently associated with inpatient mortality (Supplemental Table 1). Specifically, compared to patients with CKD Stage 1 or 2, patients with CKD Stage 5 (aOR 8.62 [95% CI: 1.82, 40.73]) had a significantly increased risk of death. A similar trend was seen for patients with CKD Stage 3 (aOR 3.80 [95% CI: 0.97, 14.81]) and 4 (aOR 4.28 [95% CI: 0.78, 23.48]), but this association was not significant. Age, ejection fraction, insurance status, and the presence of comorbid hypertension, diabetes, or atrial fibrillation were not associated with an increased risk of inpatient mortality.

In adjusted Cox regressions (Supplemental Table 1), patients with CKD Stage 5 had an increased hazard of death during the post-hospital period (aHR 2.56 [95% CI: 1.09, 6.00]). There were no statistically significant differences in the risk of death for patients based upon insurance status (Figure 2-Panel A), those with an LVEF <40% (Figure 2-Panel B), CKD, anemia or comorbid hypertension, diabetes, or atrial fibrillation. In sensitivity analyses where different outcomes were considered for participants who were out of care and not reachable by phone (Supplemental Table 2; Model 2 and Model 3), there were no qualitative differences in the findings. Similarly, in sensitivity analyses where participants with missing data were removed from the analysis (Supplemental Table 3) or where missing data were assumed to be abnormal (Supplemental Table 4), there were no qualitative differences in the findings. In Figure 2-Panel D, we display post-discharge outcomes compared to data from a prior study that examined post-hospital mortality among HF patients at BMC in 2013 (12). Over the 10 years separating the two cohorts, there was no improvement in post-discharge survival (log-rank p-value 0.33).

Figure 2

Gaps in quality heart failure care and their association with post-hospital survival

We identified significant gaps at each step of guideline-directed HF care for patients who survived to discharge (Figure 3), of which post-discharge follow-up and retention in care three months after discharge were associated with improvements in post-hospital survival (Table 2). In Step 1, 79.2% of patients admitted with HF underwent the minimal recommended testing prior to discharge. Receiving recommended testing prior to discharge was not associated with a change in post-hospital survival (aHR 1.27 [95% CI: 0.66, 2.41]). In Step 2, 50% of patients with HFrEF were on appropriate GDMT at the time of discharge; this was not associated with a change in post-discharge survival (aHR 0.98 [95% CI: 0.42, 2.30]).

Figure 3

In Step 3, 67.6% of those who survived at least one month after discharge attended their scheduled post-hospital follow-up appointment. In an adjusted Cox regression analysis restricted to patients who survived at least one month after discharge, those who attended their appointment had a substantial improvement in survival between 30 days and 31 months after discharge (aHR 0.32 [95% CI: 0.17, 0.63]; Figure 2-Panel C). In Step 4, 51.6% of those who survived three months after discharge were retained in care and had attended every scheduled outpatient appointment. Retention in care for the first three months after discharge was also associated with a significant improvement in mortality between 90 days and 31 months after discharge (aHR 0.36 [95% CI 0.16, 0.81]). In Step 5, 39.1% of HFrEF patients retained in care for three months after discharge were on GDMT. Finally, in Step 6, 16.3% of patients retained in care for three months had repeat measurement of serum creatinine and potassium after hospital discharge. The survival benefit associated with attending an early post-discharge follow-up appointment and remaining in care for three months was consistent in pre-defined sensitivity analyses (Table 2) that considered different possible outcomes for patients who were out of care and not reachable by phone, highlighting the strength and consistency of this finding across models.

Utilization of GDMT

Among the 154 patients who survived to hospital discharge, 76 (49.5%) had HFrEF. During the study period, these 76 patients discharged with HFrEF had 60 readmissions and 453 clinic visits. The utilization of GDMT for patients with HFrEF at both hospital discharge and outpatient clinic visits was low (Figure 4), with the largest gaps observed in the use of beta-blockers and MRAs. Overall, patients were more likely to be prescribed all three classes of GDMT at outpatient clinic appointments (46.6% of all visits; 211 out of 453; Figure 4b) than at hospital discharge (28.7% of all discharges; 39 out of 136; Figure 4a). Importantly, and unexpectedly, self-pay patients were slightly more likely to receive GDMT than patients with insurance, with one notable exception: insured patients received beta-blockers in outpatient clinics (79% of visits) much more frequently than self-pay patients (51% of visits). Overall, the proportion of patients on all three classes of evidence-based therapies was low.

Figure 4a

Figure 4b

Figure 5 demonstrates patterns of GDMT utilization among the 76 patients with HFrEF who survived to discharge. Each row in Figure 5 represents an individual patient, and each column represents a sequential month after index hospitalization. In the first column, which represents discharge from the index hospitalization, only 50% of patients were on all three classes of recommended GDMT; however, almost all the patients were on at least one HF medication. Similar to what was observed in Table 2 and Figure 2-Panel C, patients at the bottom of the heat map who had better care engagement were most likely to survive, even though month-to-month changes in GDMT occurred frequently, and most patients were often on fewer classes of medications as time went on. No patients received all three classes of GDMT (green in each column) throughout the duration of follow-up. Patients at the top of the figure, who died during follow-up, generally had little to no post-hospital care engagement.

Figure 5