Factors influencing survival outcomes in patients with stroke in Zimbabwe: A 12-month longitudinal study

Study design

Our study employed a longitudinal design that initially enrolled 188 stroke patients at baseline. However, during the course of the 12-month follow-up, 51 patients were excluded from the analysis due to missing information or loss to follow-up. The exclusion of these 51 observations was meant to enhance the robustness and reliability of the analysis by focusing on participants with complete and reliable data. The subsequent analysis involved the remaining 137 patients who were followed-up for 12 months, from baseline, after which the surviving participants were censored. The dataset was derived from a larger study encompassing stroke patients initiating treatment at three referral hospitals (Parirenyatwa, Sally Mugabe, and Chitungwiza Central Hospital), in Harare, Zimbabwe. At all the three hospitals, the study participants were admitted in the stroke units or medical wards. All the hospitals were manned by neurologists, physiotherapists and occupational therapists who ran weekly outpatient stroke clinics for all stroke patients discharged from hospitals.

Inclusion and exclusion criteria

All patients (age > 18 years) diagnosed with stroke and admitted to the stroke units or the medical wards of the three public hospitals during the study period were included. The study excluded admitted patients who died before medical stabilization or refused to participate in the study as well as those had a Transient Ischaemic Attack (TIA).

Study population

The source population was all stroke patients with a clinical diagnosis admitted to the three public referral hospitals in Harare, Zimbabwe from July 2015 to Nov 2017. Not all the patients had confirmatory diagnosis of stroke using a CT scan since some could not afford the cost. One hundred and eighty-eight adult stroke patients were included in the study.

Study variables

The primary outcome variable was the duration of survival until death, measured in days from the initiation of stroke treatment to the occurrence of death or censorship. Various independent variables were examined for their potential influence on the survival time of stroke patients, including sex, hospital of admission, age group, marital status, occupational status, level of education, housing status, cigarette smoking status, hospital bill payment modality, transport modality, stroke side, comorbidities, time of stroke, total functional outcome, and dependency. This comprehensive set of variables allowed for a thorough exploration of factors contributing to the survival outcomes in the study population.

Operational definitions

Event: death recorded among stroke patients during 12 months of follow-up (in the hospital or after hospital discharge in the period reported by a caregiver).

Censored: (i) if the patient was alive until the 365-day post stroke, (ii) if the patient died before they were medically stable (less than 72 hours post admission).

Time to event: Time from hospital admission until the death of the patient was confirmed either during admission or after discharge within the 12 months of follow-up.

Twelve months case fatality rate: was calculated using the number of deaths due to stroke during the 12- months follow-up period as the numerator and the total number of stroke patients during the 12 months follow-up period as the denominator

Data collection procedure

Ethical clearance was obtained from The Parirenyatwa and University of Zimbabwe Joint Research Ethics Committee (JREC – 312/12) and The Medical Research Council of Zimbabwe (MRCZ – 34/78). All participants gave written informed consent and caregivers gave assent for those that could not communicate. Data collection was carried out by one research assistant for each hospital who was trained in the use of the data collection tools. The research assistants collected all relevant data from patients’ charts and interviewed the patients/caregivers using a validated data extraction sheet and questionnaire. Participant history was recorded from the patient and/or relatives using a language they were comfortable communicating in (Shona or English). Clinical information date was abstracted from the patients’ notes or files, Data on socio- demographics (age, sex, education level occupation, marital status etc), clinical data (clinical presentation, past medical illness, duration of symptoms, CT-scan etc) was collected by interview at baseline, while death was recorded upon follow up at 3 months or at 12 months. Finally, the participants were assessed for their overall functional outcomes using the Functional Independence Measure (FIM at baseline and at follow up. This was transformed into ordinal categories (mild, moderate, severe, and very severe) to generate proportions in each category. Fifty-one participants were lost to follow up or missing information, hence data for 137 participants were analyzed.

Data analysis

The statistical analysis was conducted using STATA software version 15.1 (StataCorp, College Station, TX). Descriptive statistics were expressed as frequencies with corresponding percentages. The Kaplan Meier survival curves were utilized to compare patients by covariates. Bivariate Cox proportional hazards regression models were fitted for each explanatory variable. Variables with p≤0.25 in the bivariate Cox proportional hazards analysis were deemed eligible for further analysis. However, the Cox proportional hazards model failed the model diagnostics, as indicated by the global test result’s significance (p<0.05) and a systematic departure from a horizontal line in the plot of scaled Schoenfeld residuals against transformed time, revealing violations of proportional hazard assumptions. Consequently, Gompertz parametric regression analysis was employed for modeling the data post accelerated failure time (AFT) model diagnostics. Variables with a p-value < 0.05 were considered significantly associated with the outcome in the multivariate analysis, and the adjusted hazard ratios (HR) with a 95% confidence interval (CI) were utilized to express the associations between the outcome and independent variables. The Wald test was used to retain variables that had a p-value < 0.05, utilizing a process of iteratively removing, refitting, and verifying until all significant variables were retained in the final main effects model.

Accelerated Failure Time Model Selection and Diagnostics

The AFT models were used since the Cox proportional hazards (PH) assumptions were violated. For an in- depth exploration of the impact of candidate covariates on the survival time of stroke patients, bivariate analysis was conducted for each covariate using different baseline distributions in AFT models. The multivariable analysis incorporated the Exponential, Gompertz, Weibull, and log-normal distributions for the baseline hazard function, considering the most significant covariates. To assess the goodness of fit, estimated cumulative hazard plots were plotted against Cox-Snell residuals, providing insight into how well the AFT models captured the stroke dataset. Furthermore, the Akaike’s information criterion (AIC) was employed to assess the model’s goodness of fit and guide the selection of the most appropriate AFT model for the analysis.

Ethics approval and participant consent

Permission to carry out this study was granted by the institutional review boards at the three referral hospitals Parirenyatwa, Sally Mugabe (formerly Harare Central) and Chitungwiza Central hospitals. Ethical approval was given by the Joint Parirenyatwa Group of Hospitals and University of Zimbabwe Research Ethics Committee (JREC – 312/12) and The Medical Research Council of Zimbabwe (MRCZ – 34/78). All participants gave written informed consent and in the event that the participant could not communicate, assent was given by the caregiver.

Sociodemographic characteristics

Table 1 presents the sociodemographic and clinical characteristics of stroke patients followed for 12 months. In total 79 (58%) participants were deceased and 58(42%) were censored. Sex distribution showed 83(61.3%) females and 54(38.7%) males, with corresponding deceased percentages of 57.1% and 58.5% for each sex. Hospital of admission showed variations, with Parirenyatwa having 33/47 (70.2%) of its patients deceased, Sally Mugabe Hospital with 22/54 (40.7%), and Chitungwiza Central Hospital with 24/36 (66.7%). Age group categorization revealed higher mortality in those over 65 years (80.8%), and marital status showed 58.2% deceased among married individuals. The ’Not in Union’ category, which included the single, widowed, divorced, and separated individuals, had 56.9% deceased individuals. The unemployed category had 61.3% deceased, while among retirees 73.1% of the participants died. Educational attainment also displayed variation in mortality with 100.0% deceased among those with no formal education and 78.3% deceased among those with only primary school level education.

Overall Kaplan-Meier estimation of survival

The overall Kaplan-Meier estimation of survival, as illustrated in Figure 1, indicates that the overall probability of survival was above 50% during the first 100 days but declined to below 50% by day 200. Over the 12-month follow-up period from baseline, involving a total of 137 patients with stroke, 42% of individuals survived, while 58% were deceased.

• Download figure
• Open in new tab

Figure 1:

Overall Kaplan-Meier survival status among stroke patients

Kaplan-Meier Survivor Estimates of Patients with stroke

Survival estimates for patients with stroke were analyzed using the Kaplan-Meier method to compare survival times. The results revealed distinct patterns based on functional outcomes. For instance, individuals with very severe functional outcomes at baseline had the shortest survival time, while those with mild and moderate outcomes survived the longest. Further comparison by age group demonstrated that patients below 45 years showed higher survival times, and the lowest survival times were observed in patients aged over 60 years. Additionally, based on participant housing status, patients staying with relatives or parents and those residing as tenants had longer survival times compared to those with own accommodation (Figure 2).

• Download figure
• Open in new tab

Figure 2:

Kaplan-Meier survivor estimates by functional outcome, age group, hospital bill payment and housing status

AFT Model Selection

The efficiency of the AFT models was compared using the Akaike Information Criterion (AIC). Among the alternative AFT models considered (Weibull, Exponential, Log-normal), the Gompertz AFT model showed the lowest AIC value (AIC = 400.3), signifying its best fit to the stroke dataset compared to other alternatives. Consequently, the Gompertz AFT model was chosen as the most suitable model for describing the characteristics of the stroke dataset (Table 2).

AFT Models Graphical Evaluation

The adequacy of model fit was further evaluated through graphical analysis using Cox-Snell residuals. Specifically, Cox-Snell residuals were employed to assess the overall goodness of fit across the different parametric models. Figure 3 illustrates the Cox-Snell residuals plots, revealing that the Gompertz AFT model demonstrated the most favorable fit for this dataset of stroke patients. This determination was based on the observation that the plot of Cox-Snell residuals against the cumulative hazard function of residuals formed an approximately straight line through the origin at a 45-degree angle, distinguishing it as the optimal fit when compared to the Exponential, Weibull, and log-normal models.

• Download figure
• Open in new tab

Figure 3:

Estimated cumulative hazard against Cox-Snell residuals for the Exponential, Gompertz, Weibull and Lognormal models.

Factors influencing survival among stroke patients

In the multivariable Gompertz AFT model for survival status among stroke patients (Table 3), several predictors were assessed. Regarding hospital bill payment modalities, compared to cash payments, those receiving free services which include older than 65, pensioners and retirees had a significantly lower risk of death 0.4(95%CI:0.20, 0.80), implying a protective effect on survival. However, the ’Health insurance’ and ‘Social Welfare’ modalities were not statistically significantly associated with survival. Educational attainment demonstrated a significant trend, with higher educational attainment associated with lower risk of mortality. For instance, individuals with secondary school level education had adjusted hazards ratios of 0.2(95% CI: 0.04, 0.69), indicating a significantly reduced risk compared to those with no formal education. Age was also significantly associated with mortality, with individuals aged 45-65years and those aged over 65 years showing adjusted hazards ratios of 4,9 (95%CI:1.80,13.25) and 5.5 (95%CI:1.92,15.95), respectively, compared to those below 45 years. Housing status showed a significantly protective effect for those staying with parents/relatives [0.4(95%CI: 0.20, 0.66)], and total functional outcome demonstrated a significantly increased hazard for individuals with a ’Very severe’ outcome [4.9 (95% CI: 1.12, 21.3)].

Strengths and limitations of the study and or findings

The strengths of the study were: recruitment was from central hospitals improved generalisability, Prospective cohort design, a fairly large baseline sample size, among others. The findings of the present study highlighted the factors associated with mortality in this group of patients and paves the way for health professionals to develop guidelines to mitigate mortality among stroke patients. The limitations are attributed to missing information about the cause of death and inability to get the information thereby introducing potential selection bias despite efforts to minimize it. This led to analysis only for the available data with a substantially diminished sample size thus statistical analysis on causality could not be carried out. Our study was only on patients who were admitted in the hospitals and did not include those who died before admission and those who were not admitted due to various reasons. Wide confidence intervals on some variables suggest a possible need for replication of the study with much larger sample size. The 12- month follow-up might have limited capturing long-term survival patterns, potentially missing late-stage events and ongoing health challenges beyond this timeframe.

Conclusions

The observed influence of hospital bill payment modalities on survival outcomes, with receiving free hospital services associated with a lower risk of death, suggesting potential avenues for healthcare policy and practice improvements. Additionally, the consistent and significant trend in educational level underscores its role as a crucial social determinant of health in the context of stroke, influencing health literacy, resource access, and adherence to medical recommendations. The critical impact of age on survival outcomes emphasizes the need for tailored care strategies for different age groups. Furthermore, the protective effect of staying with parents/relatives and the heightened mortality risk associated with ’Very severe’ functional outcomes emphasize the importance of assessing and addressing social support systems and functional impairment in stroke management. These findings highlight the nature of stroke prognosis, emphasizing the need for patient-centered approach in stroke care that considers and addresses diverse determinants impacting survival outcomes. Community based studies may give more information about case fatality rates among stroke patients in Zimbabwe.