Introduction

The purpose of this blog post is to describe the prototype model developed for CSHI/Medicaid in Alabama.  It is intended to stimulate dialogue on how the model could be developed to help understand and address some of the complex issues facing healthcare provision for Medicaid patients with the chronic diseases of diabetes, congestive heart failure, hypertension, obesity and depression.

It is hoped that the prototype will provide an important stepping stone towards the ultimate objectives of the project, which has been stated as:

• To create a scalable, extensible simulation model to help identify a small set of P4P quality measures that will help providers, payors and patient paths meet quality and cost goals. 

• In addition, the model should help demonstrate potential approaches to the state Medicaid leadership and healthcare providers in a transparent manner. For providers in particular, the model will also demonstrate the level of bonuses that will be received when performance goals are met.

Given the fact that compliance is a major issue for the Medicaid program, it has been given particular focus within the prototype model. This focus has allowed the model to include cost impact modelling of Bio-monitoring. 

Design Principles

The design of the prototype has followed the following design principles:

• Focus on the important costs drivers for Medicaid patients

• Model chronic diseases as a whole to reduce model complexity

• Link into the existing health risk score work of CSHI

• Use model structures that can be extended to cater for additional requirements.

Modelling Paradigms

A hybrid model has been created using two modelling paradigms, Agent Based Modelling (ABM) and System Dynamics (SD).  One distinct advantage of ABM is that it allows easy and flexible incorporation of heterogeneity of model entities when they are treated as agents. The behaviour of the model can be driven by the specific interactions of these heterogeneous agents – this is not possible with SD modelling. However, SD has the advantage that it can be easily used to model non-linear behaviour resulting from feedback and variable accumulation.  This feature has been used in the modelling of patient compliance.

Disease State Model

Overall Structure

The underlying assumption of the prototype model is that a patient with the chronic diseases mentioned above will progress to a more serious disease states over time. The effective delivery of healthcare through interventions can have the positive effect of slowing down the progression of the disease.  Due to the higher costs associated with more severe disease states, the slower disease progression can reduce overall healthcare costs, whilst improving the patient’s quality of health. 

Following a design review meeting it was considered useful to consider the patient population in terms of different disease states or ‘bucket’ based upon the calculated Combined Health Score (CHS).

Key Assumptions

As well as patient migration between disease states, some of the most severe patients will die as a direct result of their disease. These deceased patients will be replaced by new patients who are diagnosed with one of the chronic diseases. Currently the model assumes that the flow in of new patients is equal to the flow out of deceased patients. Depending upon the time horizon of the model this assumed may not hold true.

It is further assumed that the influx of new patients is equally distributed across each of the disease states to reflect the differing disease states of patients when they are first diagnosed. This is an interesting assumption for Medicaid and can have a significant influence on the dynamics of healthcare costs.

Progression to the next disease state is determined by the respective disease state transition time.  It is assumed that more effective healthcare provision results in a longer disease state transition time, thereby changing the disease state dynamics of the patients and hence healthcare costs.  The patients within disease states are initialised with random transition times to reflect the different degrees of progression within each disease state.

Model Data

The data required to populate the Disease State Model is the following:

• Initial number patients by disease state (units)

• Worst transition times for each disease state transition (months)

• Best transition times for each disease state transition (months)

• Average healthcare costs per patient for each disease state ($)

Patient Compliance

Effective healthcare provision depends upon a number of factors including quality of healthcare provider and patient compliance. It is assumed that Medicaid patients are less likely to be compliant with prescribed treatments or self-manage of their disease. This issue can affect the beneficial impact that interventions, offered by healthcare providers, can have upon disease progression.  Managing patient compliance for Medicaid patients becomes a real opportunity.

Overall Structure

Patient compliance is modelled as the net effect of gaining compliance and losing compliance over the time.  The simple model assumes that patients gain the stock of compliance when they have an interaction with the healthcare provider. The loss of compliance arises through forgetfulness. More frequent interactions reduce the impact of forgetfulness. 

This compliance gain is based upon the advice or instructions given after an interaction, which could be a home visit, a physician appointment or an ER / hospital visit.  The magnitude of the gain will be impacted by the level of Healthcare Literacy that the patient exhibits and severity of the disease state. The model assumes that Healthcare literacy of the patient is only influenced by face to face interaction given the quality of the healthcare provider (e.g. skills, knowledge, etc). 

The anticipated behaviour is shown below. The rise in patient compliance tends to be rapid due to the nature and circumstances of the interaction. The observed decay reflects the patient forgetting the guidance, instruction or advice based upon the ‘Time to Forget’.  It is assumed that ‘Time to forget’ will increase as the patient moves to more serious disease states.

Key Assumptions

Patient compliance is defined as the propensity of the patient to follow instructions and advice given by healthcare professionals. It is measured on a scale from 0 to 1. With a level of 1 the patient will follow all advice and instructions given absolutely. With a level of 0 the patient does not follow any advice or instructions. Each patient is assigned an initial compliance level.

Interaction frequency is the number of interactions experienced by the patient within a month. It is anticipated that the number of interactions would increase as the patient disease condition becomes more severe. A key feature of Bio-monitoring is high frequency of interaction which could be days or weeks.

Healthcare Literacy is defined as the level of healthcare knowledge required to understand the instructions and advice given by healthcare professionals such that the patient can be compliant. It is measured on a scale from 0 to 1. The level of healthcare literacy will ultimately determine the maximum level of compliance that the patient can achieve.

The model assumes that increasing Healthcare Literacy is strongly influenced by the number of visits a patient experiences either through nurse home visits, physicians appointments or ER/hospital visits. The ‘Percent Face to Face’ is used to derive the number visits from the number of interactions.

Model Data

The data required to populate the Patient Compliance sub-model is the following:

• Initial Patient Compliance (value of 0 to 1) ‡

• Initial Healthcare Literacy (value of 0 to 1) ‡

• Interaction frequency (units/month) ‡

• Average Compliance Adjustment per Interaction (value of 0 to 1) ‡

• Percent Face to Face (percent 0 to 100) ‡

• Literacy Adjustment per Visit (percent 0 to 100) ‡

• Quality of Healthcare provider (value of 0 to 1)

‡Each of these variables takes initial values for each disease state to reflect the different disease state characteristics. 

Modelling Insights

A number of observations have emerged from the modelling work to date. They have led to the following interesting questions: 

• At what stage of disease progression should bio-monitoring be best deployed economically?

• What is the impact of early diagnosis of Medicaid patients upon overall Medicaid costs?

• What is the most effective way of increasing Healthcare Literacy of Medicaid patients? How important are home visits?

Enhancement ideas

Currently the model assumes that the quality of the healthcare provider is assumed to be fully effective.  Clearly this assumption will not be true for all patient / service provider interactions. Since the patient is currently modelled as an agent within the simulation, it would be interesting to model the nurse or physician as an agent as well. Attributes that include skills, attitude towards patient could be used to model the quality effectiveness of service provision at each interaction.

It would be possible to model the patient path through the healthcare system at various disease states. This would allow a more detailed build up of the healthcare costs. Also, it would give greater granularity to the change in costs associated with new healthcare interventions. This would provide the opportunity to explore improved patient / physician relations upon the chosen patient path (e.g. patient preferring to visit ER rather than making an appointment with the physician).

Graphical Interface

The graphical interface will evolve as we start to explore the dynamics of patients and healthcare costs. The upper chart plots the healthcare costs within each disease state over time. The lower graph plots the number of patients in each disease state over time to show the switching of patients between disease states. The bar chart reflects the relative proportion of patients in each disease state as the simulation runs.

It is possible to change the proportion of patient with Bio-monitoring by using the slider bar. The slider can be adjusted before running the model. This allows the impact of Bio-monitoring upon healthcare costs to be observed. A patient can be selected at random to observe specific information that includes the level of compliance and healthcare literacy, the disease state and the bio-monitoring status. The graph plots the level of patient compliance.

The buttons at the top left side of the screen (not seen in the above screen shot) allow the user to ‘run’, ‘pause’ and ‘restart’ the model.