Correlates of time to microvascular complications among diabetes mellitus patients using parametric and non-parametric approaches: a case study of Ayder referral hospital, Ethiopia

Socio-demographic and clinical factors have been known to affect the time to microvascular complications and survival probabilities of diabetes mellitus patients. The objective of this study was to identify risk factors and estimate average survival times for the time to the development of microvascular complications of diabetic patients under follow up at Ayder referral hospital from February 2011 to January 2014. A retrospective follow up study was conducted on diabetic patients during this treatment period of 3 years. Using simple random sampling a total of 277 patients were included in the study. Cox and Weibull proportional hazards models were used to identify risk factors for the time to microvascular complications. Results showed that the prevalence of microvascular complication during the follow up period in the studied population was 42.6% and was more prevalent in women. The estimated overall median survival time was 425 days. Factors associated with increased risk of microvascular complications among the sampled diabetes mellitus patients as identified by the Weibull model were older age (HR=1.025, 95% CI: 1.008-1.043, p=0.003), female sex (HR=1.531, 95% CI: 1.063-2.206, p=0.002), Type 2 diabetes (HR=2.320, 95% CI: 1.121-4.799, p=0.023) and fasting blood sugar (FBS) (HR=1.006, 95% CI: 1.0031.008, p=0.0000). The results suggest strongly that to minimize the risk of diabetic complications it is necessary to treat blood glucose aggressively.


INTRODUCTION
Diabetes mellitus is a chronic disease characterized by hyperglycemia and by complications that include microvascular diseases of the eye and kidney and a variety of clinical neuropathies (Natan, 1993).The effects of diabetes mellitus include long term damage, dysfunction and failure of various organs.Based on the revised WHO classification which encompasses both clinical stages and aetiological types, there are three types of diabetes mellitus; Type 1, Type 2 and Type 3.
Type 2 is the most common form of diabetes and is characterized by disorders of insulin action and insulin secretion, either of which may be the predominant feature (Alberti and Zimmet, 1998).
*Corresponding author: kebedeessey@yahoo.com(regional prevalence of 9.2% in 2013 and estimated prevalence of 11.6% in 2035).The corresponding figures for Africa were 19.8 million people in 2013 (a regional prevalence of 4.9%) and 41.5 million by 2035.The highest prevalence of diabetes in the Africa Region was on the island of Reunion (15.4%), followed by Seychelles (12.1%),Gabon (10.7%) and Zimbabwe (9.7%).In Ethiopia, the report indicated an estimated prevalence of 4.36 %.
People with diabetes are at risk of developing a number of disabling and life-threatening health problems.Consistently high blood glucose levels can lead to serious diseases affecting the heart and blood vessels, eyes, kidneys, and nerves.
People with diabetes are also at increased risk of developing infections.
The long-term effects of diabetes mellitus include progressive development of the specific complications of retinopathy with potential blindness, nephropathy that may lead to renal failure, and/or neuropathy with risk of foot ulcers, amputation, Charcot joints, and features of autonomic dysfunction, including sexual dysfunction.People with diabetes are at increased risk of cardiovascular, peripheral vascular and cerebrovascular disease (Alberti and Zimmet, 1998).
A multinational, open-label, observational study where 66,726 people were enrolled from 28 countries across four continents (Asia, Africa, Europe and South America) and with type 2 diabetes who had begun using insulin, found that 53.5% had microvascular complications (Litwak et al., 2013).According to (Naiker, 2009), the prevalence of diabetic nephropathy is estimated to be 14%-16% in South Africa, 23.8% in Zambia, 12.4% in Egypt, 9% in Sudan, and 6.1% in Ethiopia.Rotimi et al. (2003) looked at the prevalence and risk factors for diabetic retinopathy and cataracts in 840 patients with type 2 diabetes and 191 spouse controls, enrolled from 7 centers in Ghana and Nigeria.In this cohort (whose mean age was 46 years), the prevalence of diabetic retinopathy was 18%.A study on the prevalence of diabetic retinopathy in Jimma University Hospital, Ethiopia, showed that the prevalence of diabetic retinopathy was 41.4% (Guadie Sharew et al., 2013).
Research on estimating the mean survival time to the development of complications on diabetic patients and the determinant factors of time to the development of complications of diabetes mellitus is important to allocate resources and increase access to community health services.In a study to identify causes of end stage renal failure among haemodialysis patients in Khartoum, Sudan (Banaga et al. (2015) found that hypertension and diabetes mellitus were the leading causes of End Stage Renal Failure (ESRF) among patients over 40 years old.In a study in Sweden, Svensson et al. (2015) reported that older age, Type 2 DM, higher Body Mass Index (BMI) at diagnosis of diabetes and poor glycaemic control were risk markers for one or more of the complications of diabetes mellitus.Liu et al. (2010) in a crosssectional hospital based survey in urban China used generalized linear models to reveal that the prevalence of chronic complications varied between cities, and significantly increased with age and duration of diagnosed diabetes.Kengne et al. (2005) found that smoking, diet high in saturated fat; hypertension, obesity, type 2 diabetes mellitus, dyslipidemia, and lack of physical exercise were identified as risk factors for cardiovascular complications of diabetes mellitus in sub-Saharan Africa.Some research works have been documented in sub Saharan Africa and Ethiopia on the prevalence and determinants of diabetes mellitus and its complications (Yeweyenhareg Feleke and Fikre Enquselassie, 2005;Dawit Worku et al., 2010;Solomon Tamiru and Fessahaye Alemseged, 2010;Hall et al., 2011;Tilahun Nigatu, 2012;Solomon Mekonnen Abebe et al., 2014;Wubareg Seifu et al., 2015).However, little work has been reported on estimating the survival time to the development of complications of diabetic patients and its determinant factors.The purpose of this study was therefore to estimate the average time to the development of microvascular complications of diabetes mellitus and its determinant factors using parametric and nonparametric methods of survival analysis.

METHODS
Retrospective follow up study design was used for this study.The cohort was followed from February

Inclusion and exclusion criteria
Patients included in this study were only type 1 and type 2 diabetes mellitus patients.Patients who were exposed to gestational diabetes mellitus were not included in the study.

Statistical analysis
The variables like Age, Sex, Body Mass Index Data management and analysis of this study was carried out using STATA12.The model that best fits the data was selected using the AIC and BIC criteria.

RESULTS
A sample of 277 (168 male and 109 female)     The plot of the survivor functions of the different groups gives only an idea as to how the survival experiences of the groups may be compared.The log-rank test for the equality of survivor functions among different groups of patients is a formal test of the differences.The results of this test show that there was significant difference in survival experience of the categories of sex, residence, foot ulcer, feeding habit, albumin level, ketone level, diabetes type, hypertension and treatment, whereas the survival experience of patients do not differ significantly in the variables TB, HIV/AIDS, parasitosis and pneumonia (Table 3).Following this, all two and three way interaction terms were checked for their importance in predicting the time to microvascular complications.
However, none of them had significant contribution.
Interpretations and conclusions will thus be based on the Weibull model.However, as it was intended to use parametric and non parametric models for the analysis, the Cox proportional hazards model with Breslow method for the treatment of tied observations, was also fitted.Being female posed an increased risk which is 53% higher than male patients for developing microvascular complications.The 95% confidence interval suggests that the risk of development of microvascular complication for females is as low as 1.063 times and as large as 2.206 times compared to males after adjusting for other variables in the model.
The risk of microvascular complication was also significantly associated with higher level of fasting blood sugar.Other factors being fixed, patients with higher fasting blood glucose had higher risk of developing complications (HR=1.006,95% CI: 1.003-1.008,p=0.000).That means the hazard of developing complications increase by 10.06 when the fasting blood glucose increases by 10mg/dl.
When compared with patients with type 1 diabetes mellitus, patients with type 2 diabetes mellitus had higher risk of developing microvascular complications.The estimated hazard of the developing complication for patients with type 2 diabetes mellitus was 2.32 times that of type 1 patients (95% CI: 1.121-4.799,p=0.023).(Tesfaye et al. 1996;Wändell 1999;Moore et al., 2009;Bansal et al., 2014;Svensson et al., 2015).This is a reasonable finding, because age itself may be an independent risk factor as aging is associated with decline in fat oxidation and insulin resistance or may have other risk factors related to aging or exposure to risk factors during their lifetime.

DISCUSSION
Another important risk factor for the development of microvascular complications in patients with type-1 and type-2 diabetes is the level of fasting blood sugar.Glucose is the driving force in microvascular complications of diabetes (Reusch, 2003).Patients with higher fasting blood sugar had higher risk of developing complications.Findings from other studies also support that higher blood glucose is related to microvascular complications (Klien, 1995;Cade, 2008;Group, 1998;Cardoso and Salles 2008;Nazimek-Siewniak et al., 2002;Ali et al., 2013; Glycemic control is the primary mediator of diabetic microvascular complications. Therefore improved blood glucose control decreases the progression of diabetic microvascular disease (Ali et al., 2013).
Sex of the patient was also found to be a significant factor for the development of microvascular complication of diabetes mellitus patients.It was found in this study that female patients have more risk of developing all the three complications than males.This is supported by a similar study conducted in middle-east and North Africa (Zabetian et al., 2013) and India (Brasad and Robinson, 1995).Some other studies reported the associations between sex of a patient and specific complications.In this regard Margaret et al. (2015) reported that women had an increased risk of incident CKD compared with men after adjustment for other variables.This may be because mothers are family caretakers and hence may have high chance of exposures and less resistance to infections.In another study in India, Bansal et al. (2014) reported that being male posed high risk for nephropathy.Additional work is needed to determine if these sex differences contribute to worse outcomes in women with diabetes.
Diabetes type of the patients (type 1 and type 2) was found to be strongly associated with development of microvascular complication in this study.Many other studies also support that the type 2 diabetes mellitus patients are more likely to develop the microvascular complications than type 1 diabetes mellitus patients (Brassrad and Robinson, 1995;Herman et al., 1998).
Many research findings across the globe have reported that any increase in body mass index (BMI) above normal weight levels is positively associated with an increased risk of being diagnosed as having complications of diabetes mellitus (Wändell, 1999;Moore et al. 2009;Zoppini et al., 2012;Litwak et al., 2013;Gray et al., 2015).However this study found no significant association between BMI and the risk of development of microvascular complication of diabetes mellitus patients.One possible explanation for this surprising result may be the study design.
Because this was as a retrospective follow-up study, all patient data could not be fully retrieved and hence some important variables that may have interactions with BMI were not recorded.In addition to this, for this study, the values of the covariates were recorded at baseline values, while repeated measurements should have been taken and analyzed.Another possibility may be the existing diagnostic facilities at the Ayder referral hospital.

CONCLUSION
The present study was aimed to determine the 2011 to January 2014.The data was collected from individual patients' chart from May to June 2014.The source of population for this study was all diabetes mellitus patients of Ayder referral hospital.Since no similar researches were found in Ethiopia, estimates of hazard ratio (HR) were taken from other countries in the region.Based on the study on "Diabetes in the Middle East and North Africa" by Zabetian et al. (2013), the hazard ratio of microvascular events in Iran was HR=2.9 with 95% confidence level compared with those without diabetes.Using these estimates, the sample size for this study with power 0.8 and 10% censored observations computed using Cox proportional hazards model was 277.The 277 sample diabetes mellitus patients for this study were selected by simple random sampling technique using the patient's medical registration number from the population of 1295.The response variable is continuous and describes the duration of time (in days) until the occurrence of microvascular complications.

(
BMI), Place of residence, HIV, Feeding Habit, Albumen, Ketone, Diabetes type, Treatment type, Hypertension, TB status, Parastosis and Pneumonia were considered as potential risk factors.Descriptive statistics were used to describe the percentage and frequency of the patients in reference to all covariates.The survival experience of the patients was assessed using Kaplan-Meier survivor function.The log rank test was used to compare the survival experiences among the different groups of subjects.The Cox PH model and four parametric models (Exponential, Weibull, Log logistic and Lognormal) were used to identify the risk factors for the time to microvascular complications.The model that best fits the data was identified using the AIC and BIC criteria.The plot of versus , where is the estimate of the Kaplan-Meier estimate of the survivor function, was checked for linearity which suggests the appropriateness of the Weibull proportional hazards model.Twotailed P value of less than 0.05 was considered as significant.The Cox proportional hazards model is used in situations where the hazard of death at a particular time depends on the values of the explanatory variables.The general Cox proportional hazards model is of the form: where , the hazard function at time for the individual, is the product of two functions.The baseline hazard, , characterizes how the hazard function changes as a function of survival time; , where are observable covariates and are estimable coefficients, characterizes how the hazard function changes as a function of covariates.The ratio of the two hazards, called the hazard ratio (HR) is The HR reveals the time rate of the happening of the event at any time during the study period, compared to the reference category.HR= 1 indicates that the covariates in the model have no effect and do not change the baseline hazard, .HR< 1 and HR>1 mean that the time rates are decreased and increased throughout the study period, respectively.The above function can be expressed in the form: which is a linear model for the logarithm of the hazard ratio.The method of partial likelihood is used to find estimates of the coefficients, .Weibull proportional hazards model plays a central role in the analysis of survival data.The probability density function of the Weibull distribution with scale parameter λ and shape parameter γ, W(λ, γ), is given by The survivor and hazard functions of a W(λ, γ) constant over time and hence is equivalent to exponential survival time.If , the hazard increases with time and if the hazard decreases as time increases.The method of maximum likelihood is used to find estimates of the parameters λ and γ.
mellitus patients, who were attending regular follow up taking the oral and injectable insulin treatment at Ayder referral hospital during February 2011 to January 2014 were included in this study.Of the 277 patients under follow up at Ayder referral hospital, 42.6% have developed microvascular complications.Seventy one (25.6%)patients have developed only one type of complication, while in 35 (12.6%) and 12 (4.33%) of the patients any two and all the three complications respectively coexisted.Regarding the place of residence, 125 (45.1%) were from rural and the rest 152 (54.9%) were from the urban areas.One hundred and sixty two (58.5%)patients were taking the injectable insulin treatment while 115 (41.5%) were taking the oral treatment.Out of the 162 patients who were taking the injectable insulin treatment, 117 (72.2%) were type 1 and the rest 45 (27.8%) were type 2 diabetes mellitus patients.Females were more susceptible to develop the complication ( years with minimum of 4 years and maximum of 85 years.The median body mass index of the patients was 21.26 with interquartile range of 4.9.The diabetes mellitus patients were followed up for a median time of 425 days with the minimum follow up time of 1 day and maximum follow up time of 1095 days.The overall estimated mean survival time was 462.6 days.The simplest and informative way to compare the survival times obtained from different groups of individuals is to plot the corresponding estimates of the two survivor functions on the same axis.The estimates of the Kaplan-Meier survival functions of DM patients for different categories of the study variables are displayed using different graphs (Figure 1).

KaplanFigure 1 .
Figure 1.Kaplan-Meier survival estimates of different groups of diabetes mellitus patients at Ayder referral hospital, Ethiopia.
assess the adequacy of the fitted model in terms the appropriateness of the Weibull model and the validity of the proportionality assumption.The value of the likelihood ratio statistic for the model was 113.91 (p=0.0000)indicating overall significance of the model.Figure 2 shows the plot of log[-log(survival probability)] versus log of survival time for the two groups of DM patients and the two categories of sex.It can be observed that in both plots the two lines are approximately straight indicating that the Weibull model is plausible for the data.Moreover the two lines are nearly parallel which means that the proportional hazards assumption is valid.

Figure 2 .
Figure 2. Graphical evaluation of Weibull model and proportional hazards assumption using the plot of ln (-ln(S (t))) versus ln (t).
Figure 3. Graphs of scaled Schoenfeld residuals of the covariates in the model.
mean survival time and identify the risk factors that are significantly associated with the risk of development of microvascular complications in diabetes mellitus patients under follow up at Ayder referral hospital, Mekelle, Ethiopia, using the method of survival analysis.The mean and median survival times were 462.6 (95% C.I= 421.8-503.5)days and 425 (95% C.I. 350.1-501) days.The results of the multivariable Weibull proportional hazards regression model showed that age, sex, fasting blood glucose level and diabetes type were significant determinants of the time to the development of microvascular complications of diabetes mellitus patients.The results suggest strongly that to minimize the risk of diabetic complications it is necessary to treat blood glucose aggressively.

Table 2 .
Table1.The distribution of important socio-demographic and clinical characteristics of diabetes mellitus patients treated at Ayder referral hospital,Ethiopia (2011Ethiopia ( -2014)).Summary statistics of continuous variables included in the study of diabetes mellitus patients at Ayder referral hospital,Ethiopia (2011Ethiopia ( -2014)).

Table 3 .
Results of the Log-rank test for the significant categorical variables of diabetes mellitus patients treated at Ayder referralHospital, Tigrai, Ethiopia.(2011-2014).
included in the final main effects model.The Cox proportional hazards model and different parametric models were fitted and compared as to which model best fits the data.As can be seen from Table4below, the Weibull model having the smallest values of AIC and BIC was the best model to fit the given data.

Table 4 .
AIC and BIC values for the different models proposed for the data on microvascular complications of diabetic patients at Ayder referral hospital, February 2011-January, 2014.

Table 5 .
Multivariable Cox proportional hazards regression analysis of microvascular complications of diabetic patients at Ayder referral hospital, February 2011-January, 2014.

Table 5 .
Multivariable Weibull proportional hazards regression Analysis of microvascular complications of diabetic patients at Ayder referral hospital, February 2011-January, 2014.