Time to death and its determinant factors of stroke patients at Gambella General Hospital, Gambella, Ethiopia

Abstract

Background and purpose

A stroke or a cerebrovascular accident is a common cause of death and a leading cause of long-term, severe disability in both developed and developing countries. The most recent global burden of disease report states that there were 11.9 million new cases of stroke worldwide; stroke accounts for nearly 1 in 8 deaths globally (12%, 6.5 million deaths) and claims a life every 5 s, making it the second most common cause of death worldwide. The goal of the study was to identify the most important factors influencing stroke patients' time to death at Gambella General Hospital.

Methods

Data was gathered from patient files in a hospital using a retrospective study methodology, spanning the period from September 2018 to September 2020. R 3.4.0 statistical software and STATA version 14.2 were used for data entry and analysis. The survival time was compared using the log-rank tests and the Kaplan–Meier survival curve. The fitness of the Cox proportional hazard model was examined.

Results

The final model that was fitted was the log-logistic AFT model. A statistically significant correlation was defined as having a p value of less than 0.05 and the accelerated factor (γ) with its 95% confidence interval was employed. Eight days was the total median death time (95% CI 6–10). Significant predictors for shortened mortality time were age (γ = 0.94; 95% CI (0.0.920–0.980), hypertension (γ = 0.63; 95% CI (0.605–0.660), and baseline complications (γ = 0.24; 95% CI (0.223–0.256).

Conclusions

The shortened timing of death was significantly predicted by age, hypertension, and baseline complications. In light of the study's findings, health administrators and caregivers should work to improve society's overall health.

Introduction

A blood vessel bursts or becomes blocked by a clot, frequently leading to a stroke, which is caused by the disruption of the brain's blood flow. Damage to the brain tissue results from cutting off the delivery of oxygen and nutrients. In both industrialized and developing nations, a stroke or cerebrovascular accident is a frequent cause of demise and a major factor in severe, long-term disability [1].

According to the most recent report on the global burden of disease, there were 11.9 million new cases of stroke worldwide. Nearly 1 in 8 deaths (12%, 6.5 million deaths) were attributed to stroke, making it the second world wide’s most common cause of death [2]. Every 2 s, a stroke occurs somewhere in the world due to the increasing burden of the disease [3]. By the end of 2030, it is predicted that stroke will have increased to 23 million new cases and 7.8 million deaths per year in the absence of a strong global public health response [4]. There are no methodologically sound stroke studies in Sub-Saharan Africa, including Ethiopia [5]. Additionally, earlier studies on stroke in Ethiopia and the rest of Africa were primarily descriptive summaries of stroke kinds, subtypes, patient risk profiles, and risk factor magnitude. In Ethiopia although, admission to the hospitals due to stroke is increased time to time. According to the latest data published in 2017, stroke deaths in Ethiopia reached 39,571 or 6.23% of total deaths [6]. Stroke deaths account for 89.82 deaths per 100,000 people when age is taken into consideration [6]. In high-income nations, the estimated age-adjusted incidence rate in 2010 was 138.9 per 100,000 person-years, while in low- and middle-income countries, it was 182.6 per 100,000 person-years [7].

Studies from 61 low-income individuals found an increase in hemorrhagic and ischemic strokes of 22% (from 5 to 30%) and 6% (from −7 to 18%), respectively [8]. Even though the exact emergency burden of stroke in Ethiopia is not known, it has been estimated to be increasing and stroke accounts for 2.5% of all hospital admissions and 13.7% of medical admissions [9]. Stroke is a common and serious condition, and treatments for it have a small effect on overall health. As of 2008, it's been stated that the incidence of stroke in emerging nations has overtaken that in industrialized nations [10].

In the twenty-first century, there has been a 42% decrease in stroke incidence in the high-income countries; stroke incidence in the low- to middle-income countries has increased by more than 100% [11]. The trend is generalized because studies suggest that the geographical variations in stroke incidence and prevalence are small. While the geographical variation of stroke incidence is small worldwide, the burden of stroke shows larger geographical variation. Unfortunately, most stroke burden is carried by the low- to middle-income countries [12]. Despite the alarming threat of stroke as a major public health problem in Ethiopia, stroke epidemiology is not well-studied in Ethiopia.

In Ethiopia, stroke is a frequent cause of mortality and morbidity from non-communicable diseases. It has been shown to be the most common neurological condition seen in Ethiopia. Like other developing countries resources for stroke care and rehabilitation are deficient (poor) in Ethiopia. Patients with stroke are often poorly managed and discharged from hospital without receiving adequate rehabilitation (cure) services. This a series implication in terms of saving the life of patients especially in poorly developed societies where hemorrhage strokes which are characterized by sever neurologic presentation are very much prevalent [13]. A study which was conducted in Addis Ababa city, types and associated factor of stroke at selected public referral hospitals in Addis Ababa; Ethiopia by [10] and another study conducted in Addis Ababa city, prevalence, nursing managements and patients outcomes among stroke patients admitted to Tikur Anbessa Specialized Hospital, Addis Ababa, Ethiopia using logistic regression model, by [14]. Logistic regression has been used extensively in studies on the prevalence and risk factors for stroke, and the Cox proportional hazards model has also been used in several studies using mortality as the endpoint [10, 14]. However Logistic regression does not account for the censoring observations. Even though a semi-parametric estimate provides more flexibility, a parametric estimate is more powerful provided the baseline hazard's form is known in advance. There is limited evidence regarding the determinants of time to death of a stroke in the current study area.

Objectives

The objectives of this were: (1) to estimate median survival time; (2) to identify determinant factors associated with stroke-related death; and (3) to compare the survival probability of stroke patients among different levels of determinant factors.

Methods

Study design

The retrospective study design was gathered for patients in the medical ward [15].

Study setting

The study was conducted from retrospective records at Gambella General Hospital; Gambella Peoples National Regional State, Southwestern part of Ethiopia from the 1st of January 2018 to the 1st of January 2022 among patients who were admitted by stroke.

Participants

Stroke patients were registered and admitted to the intensive care unit ward of Gambella General Hospital during the required period and patients for whom data for variables of interest was complete were included while out of the interval period and patients with insufficient information about one of the factor variables either in the registration book or in the card that was not included.

Variables

The response variable was the time to death of patients measured in days. The survival time of the outcome of interest (death in this study) is the duration of time considered from the start of anti-stroke treatment to the date of the patient's death or censoring (i.e., 1 = death and 0 = censoring) and the independent variables are place of residence (residence) (i.e., urban & rural), sex of patients (sex), hypertension (i.e., presence or absence), cardiac disease (presence or absence), age of patients in a year, baseline complication (presence or absence), types of stroke(ischemic, hemorrhagic, Both), diabetes mellitus (presence or absence).

Study size

The study included all stroke patients who had a follow-up between 1st of January 2018 and 1st of January 2022 and satisfied the inclusion criteria. In this study, the sample size was 203, since we have 259 stroke patients in the study setting. But from these, only 203 stroke patients satisfy the inclusion criteria.

Data collection tool and procedure

Data collection was carried out by one trained nurse and one internal medicine resident. It was secondary data that was recorded on registration charts and cards via nurses, laboratory technicians, medical doctors, and clinicians.

The hospital's registry is used to extract data from stroke patients' initial date of admission up to the date of the patient’s death or censored during the study period, the medical ward registration chart and the patient's identification cards were used to select the variables in the study by trained clinicians. The cards were prepared by the Federal Ministry of Health to be uniformly used by clinicians to early identify and document clinical and laboratory variables. Thus, the data were collected from patient follow-up records based on the variable in the study. The necessary history used for the study was taken from the patient and/or caregivers by the language they understood. To ensure the quality of data, the data abstraction tool was developed in English, translated to the local language (Amharic), and back-translated into English to check its consistency. Data on socio-demographic characteristics, and clinical characteristics of patients, including risk factors, clinical presentation, subtypes of stroke, and time to death of patients owing to stroke, were collected using the data collecting form.

Operational definitions

An event is defined as the occurrence of the death of stroke patients at the time of follow-up in the hospital.

Censored are patients who were referred to other hospitals, and discharged from the hospital, or patients who didn’t develop the event and lost to follow-up.

An incomplete patient chart refers to charts that have no date of admission and major variables.

Baselines complications refer to the most common complications of stroke are Brain edema, or swelling of the brain, and pneumonia.

Stroke subtypes Strokes can be classified into two main types: ischemic (caused by a clot in a blood vessel in the brain), or hemorrhagic (caused by a bleed in the brain).

Survival time defined as the time from the date of admission to recovery from pneumonia, determined for each participant.

Statistical methods

Data processing and analysis

The data was entered into Epidata version 3.1. Statistical analysis consists of descriptive data analysis and survival model fitting to make an inference by non-parametric model, semi-parametric Cox proportional hazard models, and parametric survival (AFT) accelerated failure time models. All inferences were conducted at a 5% significance level using R version 3.4.0 &STATA 14.2 were statistical software packages used for analysis. Kaplan Meier curves and log-rank tests were used to compare the survival experience of different categories of patients [16, 17].

Univariable Cox-proportional hazards regression model was fitted for each predictor. Those variables having a p ≤ 0.25 in the bivariable analysis were selected [18, 19]. Moreover, further variable selection was undertaken using stepwise backward variable selection proportionality hazard assumption was tested using Schoenfeld residuals tests, graphically using the log–log plot of survival, and time-dependent test using (tvc) command [20,21,22]. Log logistic AFT model was employed as the final model for the study. Acceleration factor ($\gamma$) with its 95% confidence interval and p values was used to measure the strength of association and to identify the statistically significant result. p < 0.05 was considered a statistically significant association [17, 23, 24].

Results

Of a total of 259 patient medical records, 56 were excluded because of incompleteness, and the remaining 203 were included in the final analysis. Therefore this study included a total of 203 stroke patients fulfilling the inclusion criteria in Gambella General Hospital. Summary results for covariates included in this study are presented in Table 1.

Table 1 Descriptive summary of stroke patient data set at GGH from 2018 to 2022

As shown in Table 1, a total of 259 stroke patients were treated in the hospital GGH during the study period from January 2018 to January 2022. Of the total population, this study included 203 stroke patients for whom data for variables of interest are complete. Of all 203 stroke patients, 152 (74.9%) were censored or not experienced the event and 51 (25.1%) died. The study included 113 male patients of which 56.86% were dead. Among 90 female patients in the study, 43.14% died. Of the patients included in the study, 146 were Hypertensive patients of which 72.55% died, and 57 were free from Hypertension of which 27.45% were dead. There were 82 (40.39%) urban patients of which 21 (41.18%) died, 121 (59.61%) of them were rural patients of which 30 (58.82%) were dead. Of the patients included in the study, 94 (46.31%) were Cardiac disease patients of which 29 (56.86%) were dead, 109 (53.69%) were free from Cardiac disease of which 22 (43.14%) were dead. There were 112 (55.17%) ischemic-type patients of which 25 (49.02%) were dead, 63 (31.03%) of them were hemorrhagic patients of which 23 (45.10%) were dead and 28 (13.79%) of them were both type stroke patients of which 3 (5.88%) were dead. Of the patients included in the study, 62 (30.54%) were Diabetes mellitus patients of which 19 (37.25%) were dead, 141 (69.46%) were free from Diabetes mellitus of which 32 (62.75%) were dead. Of the patients included in the study, 132 (65.02%) had baseline complications of which 37 (72.55%) were dead, and 71 (34.98%) were free from baseline complications of which 14 (27.45%) were dead. The table below shows that the mean age of the respondents was 54.49 years, with a standard deviation of 21.08 years.

Among those patients with stroke, 152 (74.87%) were censored and 51 (25.13%) died, as shown in Fig. 1.

Fig. 1

Discharge status of stroke patient’s data set at GGH from 2018 to 2022

Non-parametric survival analysis

The estimated value of the survivor function patients decreases at an increasing rate from the time of origin until 10 days and remains constant after 15 days, as shown in Fig. 2.

Fig. 2

K–M plots of the survival function of stroke patients at GGH from 2018 to 2022

Figure 3 depicts the probability of survival of patients based on hypertension, demonstrating that patients without hypertension had a longer survival time than those with hypertension this indicates that patients without hypertension have better survival experience than patients with hypertension.

Fig. 3

K–M survival plot by hypertension of stroke patients

Log-rank test

The log-rank test was used to compare survival time between categories of different predictors. Based on this test, survival time among different groups of predictors such as the presence of baseline complication of patients, hypertension and cardiac disease were significantly different in survival time at a 5% level of significance and estimated median time to death stroke patients for all observations were 8 days with 95% CI [6, 10]. The median death time of patients due to stroke varied among various categories of predictors. For example, the median death time of patients who had cardiac disease was 6 days and those who had no cardiac disease were 10 days. The median times to death of patients with past baseline complication and without baseline complication were 10 and 7 days, respectively, as shown in Table 2.

Table 2 Median time to death and log-rank test by predictors of stroke patients

The univariable and multivariable analysis result

The 1st step in the model-building process is univariable analysis. Predictors which had an association at a p value of 0.25 in univariable Cox regression were included in multivariable Cox regression.

Survival of the patients is significantly related to sex, age, diabetes mellitus, and hypertension at a 25% level of significance were selected as candidate potential variables. In the 2nd step, all selected predictors in 1st step were fitted in the proportional hazard model and candidate predictors at a 10% level of significance were chosen using the backward selection method, variables duration, history of ARTI, insurance status, and clinical presentation during admission were selected as candidate potential variables.

All selected predictors were fitted in the proportional hazard model and candidate predictors at a 10% level of significance were chosen using the backward selection method. Variables of sex, age, and hypertension were selected as candidate potential variables.

All selected variables at a 10% level of significance in the second step and the non-significant variable in the univariate analysis at a 25% level of significance were modeled together using the forward selection method the following predictors were selected at a 10% level of significance.

Age, presence of baseline complication, hypertension, and diabetes-mellitus were statistically significant at a 5% significance level and those predictors were selected as the final model. It is the best model compared to forward and backward selection methods since it has the smallest value of AIC.

Using different methods predictors of age and hypertension violate the proportional hazard assumption. Thus, we doubt the accuracy of the PH assumption and consider the AFT model for this data set.

Accelerated failure time (AFT) model

When PH assumptions were not satisfied, the parametric AFT model should be used instead of the Cox model [25].

Multivariable analysis of exponential, Weibull, log-normal and log-logistic parametric models was done using all significant predictors in the final multivariable Cox PH model at a 5% level of significance. To compare the efficiency of different models AIC and BIC was used. A model having the minimum AIC and BIC value was selected as a good model. Accordingly, from Table 3, log-logistic AFT model has (AIC = 143.58, BIC = 160.70) which is selected as a good model to fit the survival time of stroke patients data than other accelerated failure time model such as exponential, Weibull and lognormal as a baseline distribution.

Table 3 AIC, BIC and log-likelihood of the candidate parametric models

The final model results are shown as follows according to Table 4 under the log-logistic AFT model. Hypertension, baseline complication, and age of stroke patients were significant at a 5% significance level. An acceleration factor greater than one (positive coefficient) indicates extending the time to death while an acceleration factor less than one (negative coefficient) indicates shortened time to death. The output of the final log-logistic AFT model is presented in Table 4. This output showed Stroke patients with hypertensive, with baseline complications and patients who were older had significantly shortened survival times. The estimated acceleration factor for patients with hypertension is 0.63 with (95% CI 0.605 0.660). The confidence interval for the acceleration factor did not include one and the p value is small (p = 0.003). This indicates hypertensive patients have less survival time than patients who are not hypertensive. Similarly acceleration factor for patients with baseline complication was 0.24 with (95% CI 0.223 0.256) the $\gamma$ CI did not include one and the p value is small (p = 0.0023). This implied the expected survival time of stroke patients decreased by 76% for patients with baseline complication as compared to patients who have no baseline complication (reference), finally holding other factors constant in the model. Finally holding other factors constant in the model, for the age of the stroke patients for 1 year change in the age of patients the log of survival time is decreased by 0.06.

Table 4 Summary result of the final Log-logistic AFT model of stroke patients

Model diagnostic

To check whether the fitted model adequately describes the data or not two graphical methods and the Likelihood ratio test were used Adequacy of Parametric Baselines plot and CoxSnell residual plot. From Figs. 4 and 5, the Log-logistic baseline distribution plot and Cox-Snell residual plot make approximately a straight line through the origin than the rest AFT models. So this plot suggests that the Log-logistic AFT model is appropriate. Also, the likelihood ratio test in Table 5 shows that the model is significant and the log-likelihood values of the null model and the full model indicate that the model had a significant improvement after the covariates were added to the model.

Fig. 4

Log-logistic baseline distributions plot of stroke patients at GGH from 2018 to 2022

Fig. 5

Cox–Snell residuals plots of log-logistic baseline distribution

Table 5 Likelihood ratio and significance of the Log- logistic AFT model

Discussion

Stroke, also known as a cerebrovascular accident, is a prominent cause of severe, long-term impairment in both industrialized and developing nations. For the effective management of stroke patients and the development of a stroke preventive strategy, the time to death and the factors that determine it are crucial. The goal of this study was to pinpoint the variables that affected how long it took stroke victims at Gambella General Hospital to pass away. A total of 203 patients were enrolled in the study to determine the associated factors of time to death for stroke patients; of those patients, 74.9 were censored or did not experience the event, and 25.1% died. This study agrees with the study conducted by [26], that 27.2% perished, while 72.8% were censored or did not witness the tragedy. People with hypertension, baseline complications, and older ages were greater at risk for stroke than person’s without hypertension, baseline complications, and younger ages. The average time for all patients was 6 with a standard deviation of 3.2 this study agrees with a study conducted by [26].

Survival models that were parametric, semi-parametric, and nonparametric were all used in this investigation. Based on the Kaplan–Meier estimate approach, a non-parametric method is utilized to compare the differences between each categorical covariate. The Cox PH model was used to fit a semi-parametric survival analysis. Schoenfeld residuals and the Cox PH model's assumptions were tested graphically, and both were shown to be false. The researcher then proposed a parametric AFT survival model as a substitute for the Cox PH model to suit the pneumonia data from Gambella General Hospital. For the Stroke patient dataset at Gambella General Hospital, the researcher fit AFT models using several baseline distribution patterns. The baseline distributions used in this study were Exponential, Weibull, Log-normal, and Log-logistic. The log-logistic AFT model was selected as a better AFT model than Weibull, Exponential, and log-normal models based on comparison criteria with smaller AIC and BIC values. The overall median time from stroke patients was 8 days (mean = 6 days; standard deviation = 3 days). This study is almost consistent with the Research conducted by [26].

Age, baseline complications, and hypertension were statistically significant predictors of the survival status of stroke in this study. This study is consistent with the study conducted by [26] Hypertension is a highly predictor of death of stroke patients. This is consistent in the literature, multiple studies have identified hypertension as the leading risk factor for stroke in SSAs [27]. Other studies have found that hypertension is an important modifiable risk factor for stroke and hypertension was the most frequent co-morbidity that occurred in 50.6% of all stroke patients [28].

Age was found to be a key determinant in this study when determining the time until a stroke patient died; as patients aged, their chances of survival reduced. This result is consistent with another investigation in the literature that found that becoming older is the primary, non-modifiable driver of stroke risk [28]. For this study, there are no significant differences between rural and urban patients with time to death. In contrast, another study conducted in Tanzania showed there were significant differences between rural and urban populations with time to death [29].

Conclusion

This study used the survival time of Stroke patients' dataset of those patients who started their Stroke treatment from 1st January 2018 to 1st January 2022 years to determine the determinant factors of time to death of Stroke patients in Gambella General Hospital. Out of the total 203 stroke patients who started Stroke treatments, about 25.1% died at the end of the study. The estimated median survival time of stroke patients was 8 days.

To determine the associated factors of survival time of stroke patients, the Cox PH model was used and the PH assumption was checked by graphical, Schoenfeld residual plot and global test. Then, AFT model was fitted because the assumption of the Cox proportional model was violated. Different AFT models using different baseline distributions were applied. Among them using AIC and BIC, the Log-logistic AFT model is a better-fitted survival time of Stroke patients' dataset than other AFT baseline distributions.

The best model to fit the data to explain the survival time of the Stroke patient dataset in Gambella General Hospital was the Log-logistic AFT model, which was revealed using the graphical technique and Cox-Snell residuals plots.

In Gambella General Hospital, the results of a Log-logistic AFT model revealed that age, hypertension and baseline complication were found to be determinant factors of the survival status of stroke patients. Patients without hypertension and baseline complications had considerably longer survival time (higher survival experience). While 1-year increases in age (older age) shortened the survival time by 0.94 times. The health giver to be planned and awareness about the risk factors of stroke, and the benefit of regular medical checkups and treatment follow up should be given to the community.

Strategies for screening and management of hypertension, age and baseline complication should be given priority as they are the most prevalent determinant factors identified. Identifying and managing early stroke complications are important for the prevention of early stroke related mortality. To prevent strokes we should focus on reducing vascular risk factors such as high blood pressure stroke patients.

Based on this study, the following recommendations are forwarded for policy makers and the responsible bodies: age, baseline complication and hypertension were significant factors and need to be considered when planning and developing policies against stroke to increase patient’s survival time. Additionally, special attention should be given for old age patients in order to prolonged death timing.

Based on the finding of the study the following recommendations were made for ministry of health, the community at large, Gambella General Hospital and researcher. Community outreach program has to be planned and awareness about risk factors of stroke, benefit of regular medical checkup and treatment follow up should be given to the community and periodic follow up and adherence to the treatment of determinant hypertension, baseline complication can minimize the chance of getting stroke.

Limitation of study

The following were some of the study's limitations. There is a lack of published literature in the country regarding the survival time of stroke disease, with references to the outcomes of other countries. This study used Gambella General Hospital data from a single hospital, which does not represent the prevalence at the national level, but the patients came from different regions of the country. As the data is gathered from the treatment card of patients of the study has limited number of variables considered as risk factors for the survival time of stroke patients.