DIABETES PREDICTION
With success in modeling for the infectious diseases as mentioned earlier in Section 2: Introduction, researchers began to work with model development for non-communicable or ‘lifestyle’ diseases/disorders. ‘Markov Models’ are the models of choice in this regard because they take into consideration the current use of resources and the outcomes and are not dependent upon the historical data. Markov models are useful when a decision problem involves risk that is continuous over time, when the timing of event is important, and when important event may happen more than once. Representing clinical settings with conventional decision trees is difficult. Markov models assume that a patient is always in one of a finite number of discrete health states, called as ‘Markov States’. All events are represented as transitions from one state to another \cite{Sonnenberg_1993}. One of the foremost examples of a ‘lifestyle’ disorder is diabetes and the proportion of afflicted individuals worldwide is increasing at an alarming rate. It is now standing as one of the top 10 leading causes of deaths occurring worldwide. As per the latest WHO report (WHO 2016), the number of diabetic people rose to 8.5% in 2014 from 4.7% in 1980s and the increase is observed across the populations of all age groups. Diabetes leads to several premature deaths because of its associated adverse complications like cardio-vascular, renal and neuronal disorders. Between the two types of diabetes - 1 and 2 - the latter affects majority of the population and worryingly the onset age is decreasing. Furthermore, diabetes is a costly disease accounting for between 2.5 and 15% of the total healthcare expenditure \cite{Boutayeb_2004}. International Diabetes Federation estimates that globally 415 million people were afflicted with diabetes in 2015, and the number is ‘predicted’ to increase to 642 million by 2040. Among the top 8 countries (having highest number of diabetic patients) in the world, India stands second with 65.1 million people with diabetes and another 36.5 million with pre-diabetes (a high-risk condition for diabetes and cardio-vascular disease) (Table 4).
This burden continues to increase, driven by nutrition, lifestyle and demographic transitions, unhealthy dietary habits, and physical inactivity, in the context of a stronger genetic predisposition to diabetes \cite{Tripathy_2018}. The cost of diabetic medications is increasing in both urban and rural India. Thus, it would be really useful if with the help of modelling, both the number of diabetic patients over the future years and the costs incurred on them over this time could be reasonably predicted.
There are two main types of diabetes:
1. Type 1 Diabetes or Insulin Dependent Diabetes Mellitus (IDDM) develops if the pancreas is unable to produce insulin. This type of diabetes usually develops before the age of 40 and is commonly treated by insulin injections and a controlled diet.
2. Type 2 Diabetes or Non-Insulin Dependent Diabetes Mellitus (NIDDM) develops when the body cannot generate enough insulin (Hypoinsulinaemia) or if the body cannot use the insulin generated efficiently (Insulin Resistance). This condition is commonly treated by diet and exercise alone or by diet, exercise and tablets and/or insulin injections (WHO, 2016).
One of the earlier modelling studies was reported from Aintree Hospital, UK \cite{2003a}. Based on earlier years of available data for diabetic patient hospital admission, health records and subsequent progress, ‘Regression Analysis’ was performed to estimate various parameters for the year 2030. Attributes such as NIDDM prevalence (3035 new cases predicted), risk factors (increased obesity and other unknown contributory factors), burden on health services and monetary costs were predicted (>46.6%). Other prominent UK based studies, detailing cardiovascular risks of diabetic patients, are reported by UK Prospective Diabetes Study (UKPDS) groups\cite{Stevens2001,Kothari2002,Clarke2004}. \cite{Boutayeb_2004} developed a simple model based on differential equations, to monitor the size of the diabetic population and to monitor the transition from the stage of diabetes without complications (D group) to the stage of diabetes with complications (C group). This model is popularly known as the ‘DC Model’. Here the number of new patients (incidence) of diabetes is assumed to be constant over time. The paper stressed upon the importance of early healthcare interventions before progression of diabetes towards complications. But soon it became evident that different from study by Boutayeb, the numbers of incidences of diabetes are not constant \cite{Boutayeb_2004}. They vary based on lifestyle factors and genetic factors. Hence, the DC model was recently evolved into ‘Susceptible Diabetes Complication (SDC) Model’ \cite{Widyaningsih_2018}. Here besides the D & C groups mentioned in the DC model, an additional group of susceptible individuals (S) is created. The model predicted that diabetes prevalence is supposed to increase to 14% by 2030 and the death rate is predicted to increase by 5% every year. A similar study to UK study mentioned above, was carried out in USA \cite{Huang_2009} but used the ‘Markov Model’. Here both the number of patients and the costs incurred were predicted. The study predicted 44.1 million diabetes patients by 2034 and the medical spending on them to be USD 336 billion. The study also extended upon the UKPDS model by considering the glucose concentration data. The predictions follow the same trend as the UK study: increase in number of diabetic patients with increasing obesity and the increased medical cost incurred by the patients as well as the government.
Currently, following predictive modelling tools (Table 5) are available for disease data but their discussion in detail is outside the context of the present review.