What is the background of the project?
Diabetes is a major health problem all over the world that leads to severe complications and disability. South Asia is one of the most affected regions and with a prevalence of around 9.7 %. Bangladesh is the country with the second-largest number of adults with diabetes in South Asia, showing an increasing trend in both urban and rural areas (Safita, et al., 2016). However, awareness of diabetes among patients in Bangladesh is still poor (Saleh et al., 2012).
Type 2 diabetes (T2D) Mellitus, also known as adult-onset diabetes mellitus or noninsulin-dependent diabetes mellitus, is a serious chronic illness and depression is another major public health concern (Regier et al., 1993). Although epidemiological data (Ali et al., 2006; Mezuk et al., 2008) suggest that depression may increase the risk of diabetes onset. However, the nature and directionality of the association between depressive symptoms and subsequent diabetes course remain unclear. This is because cross-sectional studies generally indicate modest contemporaneous associations between depression and both glycemic control and complication rate but cannot speak to directionality. Most diabetes treatment trials overlook depression assessment or include insufficient numbers of depressed patients to permit subgroup analyses. With these exceptions, longitudinal studies relating T2D course to temporal variation in depressive symptoms are rare.
To determine the cause, manifestation, and outcome of wellness and disease, George L. Engle developed the biopsychosocial model of health and illness that states the interactions between biological, psychological and social factors (Engel, 1977). This model is for clinical practice and research for psychologists, nurses, physicians, and social workers (Hatala, 2012). According to Engel’s model any disease such as depression is caused by biological (physiological or genetic predispositions), psychological (health beliefs and lifestyle) and social factors (family relationships, socioeconomic status, and social support) (Habtewold et al., 2016; Garcia-Toro & Aguirre, 2007). The model reveals the interaction of this factor to create a patient’s state of mind and body (Havelka et al., 2009). However, no study has found to identify biopsychosocial risk factors of depressive symptoms in diabetic patients in Bangladesh.
What techniques and methods are used?
The expected techniques and methods of the study are given below:
Primary Data: Sample size and sampling plan
To achieve the research objectives, the study will conduct a cohort study of registered diabetes patients in Sylhet city corporation.
Sylhet diabetes hospital is the only specialized hospital where the patients are registered with diabetes. From the registered diabetes patients, the sample will be selected who tested the glycated hemoglobin (HbA1c) level at the time of data collection — assuming that most of the registered diabetes patients have been visited by the hospital above or any of the well-known government hospital and private clinic (with continuous access).
Three inclusion criteria will be applied to select the patients: T2D diagnosis at least for one year, age ≥25 years old, capable of independent communication, and signed written informed consent. Patients have already treated for depression, or other psychological illnesses (e.g., anxiety or personality disorders) will be excluded.
Sample size calculation
The minimum required sample size will be calculated using the following formula
C=Effect Size=1/2 ln[(1+r)/(1-r)]=0.32, r is the correlation be¬tween the depression and quality of life (r = 0.31) based on the studies (Eren et al. 2008; Derakhshanpour et al. 2015; Saleh et al. 2015),
Z_(1-α⁄2)=1.96, α=0.05 is the level of significance,
Z_(1-β)=0.84, 1-β = 0.80 is the power of the test,
p_0=0.20 is the known prevalence i.e., 20% of patients who were aged >25 years having diabetes for at least 1 year (from the date of interview) (Saleh et al. 2015).
Sample size estimation determines the number of complete cases which are needed for analysis. However, some subjects who enroll in the study may drop out, others may be protocol failures and still, others may have incomplete data, especially on the key outcome variables. Since our study is the longitudinal study, the sample sizes should also be computed with attention to dropout rates. In that case, attrition rate must be considered to attain the expected number of subjects.
To deal with this, decide on an attrition rate and inflate the sample size by the following factor. In our study, we expect to lose about 15% of the sample after three years of follow-up, then the sample size should be increased by a factor of
AF =1 / (1 - 0.15) ≈ 1.18.
That is, we will enroll 18% more subjects that the sample size calculation called for. Therefore, the final sample size in our study will be
Systematic random sampling technique will be used to reach individual patients. At baseline (time point-1), the data will be collected by trained personnel from every three patients (sampling interval, K = N/n = 3), in the way that the first patient will be randomly selected. Then, a number will be selected from 1 to 3 (the second patient will be selected). All biological (physiological) data will be collected from patients’ medical chart/prescription. The face-to-face interview will be conducted to collect psychological and social data.
At a 15% attrition rate taking into account, the total number of baseline sample (n = 472) will be followed-up next three years with an interval of 6±2 months. The minimum and maximum visit will be 4 and 5 (optional) because of budget and time constraints.
An interviewer-administered questionnaire will be designed to know and assess the level of the patient’s information on age, sex, educational qualification, occupation, monthly income, duration of diabetes, family history of diabetes, acquisition of information relating to diabetes, and prescribed treatment for patients and other biopsychosocial factors will be collected by interviewing the patients. A checklist will be used for collecting HbA1c data from the patients’ guidebook. The checklist means an instrument used when observing some situation. The researcher/interviewers put tick marks against the particular point or wrote down what he/she observed.
The term HbA1c refers to glycated hemoglobin. It develops when hemoglobin, a protein within red blood cells that carries oxygen throughout the body, joins with glucose in the blood, becoming 'glycated'. By measuring glycated hemoglobin (HbA1c), clinicians are able to get an overall picture of what our average blood sugar levels have been over a period of weeks/months. HbA1c is a term commonly used in relation to diabetes.
Measuring depressive symptoms
The Patient Health Questionnaire (PHQ-9) is the depression module, which refers to symptoms experienced during the last two weeks (Yu et al., 2012). It includes nine items with individual score ranges from 0 (not at all) to 3 (nearly every day). The total sum score ranging from 0 to 27. Using the PHQ-9 scores, depression severity measures with a cut-off point: 0-4 none, 5-9 mild, 10-14 moderate, 15-19 moderately severe, 20-27 severe (Kroenke et al., 2001). The validity of PHQ-9 has been assessed against an independent structured mental health professional (MHP) interview. PHQ-9 score ≥10 had a sensitivity of 88% and a specificity of 88% for major depression (Kroenke et al., 2001). In our study, T2D patients’ depression status will be measured by administering a validated Bengali version PHQ-9 questionnaire.
Health-Related Quality of Life
The demographic questionnaire will be completed in the presence of patients, and related data to the disease and its complications will be obtained from their records. Patient Health Questionnaire (PHQ-9) will be used to assess depression, and the short version (26 items) of the World Health Organization Quality of Life Questionnaire (WHOQOL-BREF) will be used to assess the quality of life. In the 26-item WHOQOL-BREF, the first two items assess general health status and quality of life, and the next 24 items assess the quality of life in 4 domains of physical health, psychological health, social relationships, and environment using 3, 6, 7 and 8 items, respectively. The 5-point Likert scale will be applied for items 1 and 15 (very bad, bad, not bad, good, and very good), items 2, and 16 to 25 (very unhappy, unhappy, not unhappy, happy, and very happy), items 3 to 14 (not at all, a little, medium, high, and very high), and for item 26 (never, rarely, occasionally, often, and always). Scoring for items 3, 4, and 26 is in reverse. To compare domains and the first two items in depressed and non-depressed groups of patients, quality of life score for each patient (Xi) is converted into scores between 0 and 100 using the following equation (Nedjat et al., 2008; Vakili et al., 2012):
Validity and reliability of WHOQOL-BREF have been reported favorable in many native and foreign studies (Alonso et al., 2004; Sadat et Al., 2014; Skevington et al., 2004). Derakhshanpour et al. (2015) estimated the validity of this questionnaire according to internal consistency for the whole scale of 85%, physical health 78%, psychological health 80%, social relationships 69%, and environment 84%.
Biopsychosocial risk factors will be found to characterize the depressive symptoms level using cross-sectional data by applying multiple linear regression and Mediation analysis will be conducted to find out the mediation effect of HbA1c. Briefly,
Multiple Linear Regression Model
The regression analysis will begin by considering a model containing all covariates (both quantitative and qualitative). Cohen (1968) and Hardy (1993) proposed that any combination of categorical and continuous variables can be analyzed within a multiple regression model framework simply through the dummy coding of the categorical variables. In general, a first-order multiple regression model is of the form:
Where, Yi = continuous response variable (depressive symptom scores) for the ith observation,
β0, β1,…, βp-1 = partial regression coefficients which measure the magnitude of the association of the X’s with Y,
Xi1, Xi2… Xip-1 = the known constants, namely, the values of the predictor variables of the ith observation,
εi = a random error term which is assumed to be independent and normally distributed with the mean zero and variance σ2, COV(εi, εj)=0, i ≠ j.
The above procedure will be conducted to verify the objective-1 and objective-2.
The linear mixed model will be applied to find out the longitudinal association between depressive symptoms and quality of life.
Linear mixed models
Repeated measures on subjects are very common in health, social, behavioral and biological sciences. The major challenge in analyzing repeated-measures data is the fact that the measurements on a subject are correlated. This correlation must be taken into account during the statistical analysis to obtain valid inference, i.e. estimates of effect sizes for the association between parameters and outcomes of interest. The correlation can often be captured by introducing random effects in the classical statistical analyses, e.g. linear and logistic regression. These statistical models combine the components of fixed effects, random effects (e.g., random-intercept and random-slope), and repeated measurements in a single unified approach. These models are called linear mixed models (LMM) for continuous longitudinal outcomes (Laird and Ware, 1982; Verbeke and Molenberghs, 2000), and generalized linear mixed models (GLMM) for other type of outcomes (Breslow and Clayton, 1993; Molenberghs and Verbeke, 2006).
The mathematical equations for LMM are explained briefly here for longitudinal data and subject-specific time profiles. Let Y_it be the response measure (e.g. qu,ality of life and depressive symptoms) for subject i (=1, 2, …, N) measured at time the T_it, t = 1, 2, …, n_i. The linear mixed model is simply defined in the matrix form as
Y_i=X_i β+Z_i b_i+ε_i
Where, b_i~N(0,D), ε_i~N(0,Σ_i), Y_i is the n_i-dimensional response vector for subject i, X_i and Z_i are the (n_i×p) and (n_i×q) design matrices of known covariates, β is the p-dimensional vector of population-average regression coefficients (fixed effects), b_i is the q-dimensional vector of random effects for subject i, ε_i is a n_i-dimensional vector of measurement error components. It is assumed that b_i and ε_i are independent. Conditional on the random effects b_i, the distribution of Y_i is given by Y_i |b_i~ N(X_i β+Z_i b_i,Σ_i). The inference is based on maximizing the likelihood function of the marginal response Y_i. More specifically, a subject-specific time trajectory of polynomial form can be defined as
Y_it=β_0+∑_(k=1)^p▒∑_(r=0)^q▒〖(β_rk X_ik+b_ri )*T_it^r 〗+ε_it
where, β_0 is the overall intercept, β_rk are the rth (r = 0, 1, 2, …, q) polynomial form of time and kth (r = 1, 2, …, p) fixed effects parameters, b_0i is the random-intercept, b_ri is the random-term for time ordea r T_it^r of subject i, and ε_it is error disturbance term. The model parameters and variance components are estimated by either Maximum Likelihood (ML) or Restricted Maximum Likelihood (REML) estimation procedure (Verbeke and Molenberghs, 2000). Note that each subject has its own time profile in this model.
The linear mixed model (LMM) will be conducted to verify the objective-3 to explore the clinical heterogeneity.