MobiGuide: a personalized and patient-centric decision-support system and its evaluation in the atrial fibrillation and gestational diabetes domains

Abstract

MobiGuide is a ubiquitous, distributed and personalized evidence-based decision-support system (DSS) used by patients and their care providers. Its central DSS applies computer-interpretable clinical guidelines (CIGs) to provide real-time patient-specific and personalized recommendations by matching CIG knowledge with a highly-adaptive patient model, the parameters of which are stored in a personal health record (PHR). The PHR integrates data from hospital medical records, mobile biosensors, data entered by patients, and recommendations and abstractions output by the DSS. CIGs are customized to consider the patients’ psycho-social context and their preferences; shared decision making is supported via decision trees instantiated with patient utilities. The central DSS “projects” personalized CIG-knowledge to a mobile DSS operating on the patients’ smart phones that applies that knowledge locally. In this paper we explain the knowledge elicitation and specification methodologies that we have developed for making CIGs patient-centered and enabling their personalization. We then demonstrate feasibility, in two very different clinical domains, and two different geographic sites, as part of a multi-national feasibility study, of the full architecture that we have designed and implemented. We analyze usage patterns and opinions collected via questionnaires of the 10 atrial fibrillation (AF) and 20 gestational diabetes mellitus (GDM) patients and their care providers. The analysis is guided by three hypotheses concerning the effect of the personal patient model on patients and clinicians’ behavior and on patients’ satisfaction. The results demonstrate the sustainable usage of the system by patients and their care providers and patients’ satisfaction, which stems mostly from their increased sense of safety. The system has affected the behavior of clinicians, which have inspected the patients’ models between scheduled visits, resulting in change of diagnosis for two of the ten AF patients and anticipated change in therapy for eleven of the twenty GDM patients.

Appendices

Appendix 1: Care provider first interview: psycho-social and demographic concepts elicitation

The questionnaire below refers to the personal information that is gathered during the patient interview (history taking), work up, treatment and follow-up treatments. The medical information of a patient, created during physicians-patient encounters, is usually stored in an electronic medical record (EMR) of a specific healthcare facility. The medical record is designed per the requirements of each medical facility, but nevertheless there are generic parts that are common to all medical records, such as family history or personal history (social history) that contains the personal data that affects the patient and therefore impacts treatment decisions.

This questionnaire is designed to identify psycho-social and demographic concepts. These concepts induce context that may have effects on patient management. As well, the questionnaire will help to identify these effects.

For our first part—please think of the main psycho-social and demographic concepts that have the greatest influence on treatment recommendations:

• a. What is the psycho-social or demographic concept?

• b. What is the main effect of the contexts induced by the concepts on clinical goals or recommendations?

For the second part—we will refer to a situation where a patient has entered your clinic. The patient complains that he/she suffers from chest pain; last lab exam revealed that he has high cholesterol and he/she seems over weight.

The patient-doctor encounter will usually include the following steps of data collection:

a.	Personal details	[coded options]
b.	Chief complaint [in our case – chest pain]	[coded options]
c.	Current disease [elaboration of chief complaint]	[free text]
d.	Current problem list [emphasis on relevant diseases]	[coded options]
e.	Past medical problems	[coded options]
f.	Family history.	[coded options + text]

Please describe the patient-doctor encounter, with emphasis on psycho-social and demographic concepts:

• 1. Basic psycho-social and demographic concepts, please refer to specify:

  • 1.1. What are the main psycho-social and demographic concepts you collect during the medical interview?

  • 1.2. How do you take this information into consideration during the patient workup?

  • 1.3. How does each concept affect the treatment goals and diagnosis/work up process? [sending the patient to other exams, next meetings and other considerations]

[Basic psycho-social and demographic concepts include: degree of routineness of the daily schedule and diet, degree of family support, etc.]

For the next part we will refer to a patient with chronic medical condition who requires treatment, monitoring, and follow-up.

• 2. Please think of a patient of yours who already has an established diagnosis (e.g., AF, hypertension). Can you describe the process of patient management for such a patient?

  • 2.1. Specify the major decision points and treatment goals during current line of treatment for this patient.

  • 2.2. What was the role of psycho-social and demographic context in each treatment goal decision?

  • 2.3. Based only on clinical information (ignoring the psycho-social and demographic contexts), what are the other treatment options available?

  • 2.4. Per treatment recommendation—define different psycho-social and demographic contexts that will change the treatment recommendation.

For the next part we will assume that we have a new patient guidance system that integrates hospital and monitoring data into a Personal Health Record (PHR) accessible by patients and care providers.

The system includes Decision Support System (DSS) at the back end (health center), and on the front end (patient), by utilizing monitoring technologies. The distributed DSS provides information to the patient and care providers. Customization will be achieved by considering psycho-social/demographic data.

The following description refers to the domain of gestational diabetes. Please refer to this example and create an analogous scenario from your clinical domain.

Pazit is 39 years old; she has 3 children and now expecting the 4th one. She is a manager in a small projects company. Her duty requires a lot of car driving, meetings and out of the office scheduling.

Pazit was diagnosed with Gestational Diabetes Mellitus (GDM).

Pazit has been enrolled in the new system (the MobiGuide system). She has received a Smartphone, and devices to monitor her condition. Her state requires insulin therapy, so the patient monitoring needs to be strict.

She received recommendations to monitor blood glucose levels 4–6 times per day (fasting and 3 postprandial—1 h—measurements (3 measurements taken 1 h after each meal), and possible addition of pre-lunch and pre-dinner). She is instructed to monitor insulin (doses and times), intake deviations from the established diet recommendations (daily), ketonuria (daily), and body weight (weekly). The Smartphone will monitor her physical activity according to the defined guidelines and the patient health state (in some cases the physical exercise could have medical or obstetrical contraindications, so the goal will be to monitor the physical activity to guarantee that the patient is resting).

She has access to MobiGuide via a mobile application and also to a web application.

The next visit is scheduled 2 weeks later, if there are no relevant events.

Pazit is now in her 28th week and in the next 2 days Pazit has a big event and presentation in a conference.

• 3. In the process of patient follow-up, treatment and work up, what is the information that you expect to gain from the system?

  • 3.1. Specify the decision making points in the process, indicating the reasons for selecting each decision option.

  • 3.2. What are the mobile monitoring devices that the patient can use to supply the physicians with reliable information that is indicative of her state and to what extent do you think that they are reliable (for example glucose meter, a pedometer, an accelerometer and a pulse-meter)?

  • 3.3. What is the personal information required in this monitoring process?

  • 3.4. What is the effect of each item of psycho-social or demographic data on the treatment recommendations?

  • 3.5. Per treatment recommendation—define the different psycho-social and demographic data that will change treatment recommendations.

  • 3.6. Specify data values that will generate notification or will cause you to consider rescheduling the next meeting with the patient?

• 4. For each psycho-social and demographic data item:

  • 4.1. What are the main values that the data item may take (scale)?

  • 4.2. Who should enter the information into the system (physician, nurse, other care provider or patient)?

  • 4.3. Who should be able to change the information?

  • 4.4. Would you consider the option of subscribing mechanism (e.g. RSS-like) for checking the events (clinical or non-clinical) that had happened in the last period (hour/day/week) which would trigger change in the value of a psycho-social or demographic data item?

  • 4.5. What are the specific values of psycho-social or demographic data items that require immediate notification?

    • 4.5.1 Who should get the notification (Patient/Physician/Nurse/other care provider)

• 5. In cases a monitored patient doesn’t have web connection, namely in case the patient has only the smartphone and the monitoring system (no connection option to main server):

  • 5.1. What do you expect that the system do to notify the patient (only continue monitoring? notify changes and suggest actions? Etc.)

  • 5.2. What are the decisions the system should take?

  • 5.3. What situations must be addressed immediately?

Thank you for your cooperation.

Appendix 2: Patients’ first interview: psycho-social and demographic concept elicitation

The questionnaire below refers to the personal information that is gathered during interview with the care providers (history taking), work up, treatment and follow-up treatments. The medical information that gather during those meetings is usually stored in an electronic medical record (EMR) of a specific healthcare facility.

This questionnaire is designed to identify psycho-social and demographic concepts. These concepts induce context that may have effects on patient management. As well, the questionnaire will help to identify these effects.

For our first part—we will refer to a series of treatments that you had with your care provider.

Please describe the course of the meetings, with emphasis on treatments and decisions in a time period:

• 1. What were the main psycho-social and demographic concepts to which you had referred during the meetings/s?

• 2. How did the treatments influence your daily routine, in terms of:

  • 2.1. Difficulties you encountered.

  • 2.2. Personal decision points you took, during daily treatments?

• 3. Family support:

  • 3.1. Did you have family support during the treatments?

  • 3.2. What was the level of the family member involvement in treatments, and the decisions impacted?

• 4. How did it reflect in the process (daily activities, meeting, etc.)? What habits and daily functions were changed during the treatment, and how it was reflected in the process?

• 5. Can you specify significant event$\backslash$s that made a (good$\backslash$bad) change in the treatment process or daily routine?

• 6. What is the most burdening element, why and how did you handle it?

• 7. Did the treatment influence your daily functions:

  • 7.1. How did the treatment process affect your work?

  • 7.2. How did the treatment process affect your hobbies?

  • 7.3. How did the treatment process affect your habits?

• 8. Did you change your diet, based on treatment recommendation?

  • 8.1. Did you keep the nutrition habits and for how long?

For the next part we will assume that we have a new patient guidance system that integrates hospital and monitoring data into a Personal Health Record (PHR) accessible by patients and care providers.

The system includes Decision Support System (DSS) at the back end (health center), and on the front end (patient), by utilizing monitoring technologies. The distributed DSS provides information to the patient and care providers. Personalization will be achieved by considering patient preferences and psycho-social and demographic context.

• 9. What are your expectations from the system, regarding the following aspects:

  • 9.1. Messages and type of information that you want to receive from the system?

  • 9.2. What are the main features, you think, the system should present?

  • 9.3. What new options the system will help you achieve?

• 10. How do you think the system will influence your treatment process?

• 11. How the system should help you follow treatment recommendations (reminders, medications, etc.)?

• 12. What don’t you want that the system to do?

• 13. In cases you don’t have web connection, namely in case you have only the smartphone and the monitoring system (no connection option to main server): What do you expect that the system will do? (notification, reminders, decisions support, suggest actions, etc.)?

Thank you for your cooperation.

Appendix 3: Care providers’ second interview: Refining and generalizing psycho-social and demographic concepts

We conducted preliminary qualitative studies consisting of interviews and questionnaires with different stakeholders. Based on the questionnaire analysis we defined the emerging psycho-social and demographic concepts that induce contexts which influence patient treatment recommendation and treatment goals.

Part 1—defining scale for psycho-social and demographic concepts and the potential effects of the contexts that they induce on decision-making

The following list presents general psycho-social and demographic concepts defined by physicians, nurses and patients:

Ability to comply with treatment, communication level, cooperation level, desire to know truth about prognosis, education level, language level, trust level, need for accompanying person for visits, degree of family support, degree of routine of daily schedule and diet, distance from medical center, financial capability, living area accessibility, living area pollution, and family support level.

• 1. Per each concept, please define the following information:

  • 1.1. Scale (e.g. distance from medical center = close, medium distance, far, very far away)

  • 1.2. If relevant, what are the minimum and maximum units (e.g., close = 15 km)

  • 1.3. What are the typical use cases scenarios of each concept-induced context, and can you think of personal events that induce these contexts?

• 2. Can we gather concepts into groups:

  • 2.1. Please specify the group and the concepts under it.

  • 2.2. What are the main characteristics of each group?

• 3. Per each context—What are the potential effects on treatment goals and recommendation:

  • 3.1. During physician-patient encounter (treatment goals, clinical measures, medication doses, limitations, etc.)

  • 3.2. During daily routine (clinical measures, medication doses, limitations, treatment options etc.)

Part 2 –effect of each context

• 4. For each concept—please consider all possible values. For each context corresponding to a concept value, what would be the effect on treatment recommendations?

• 5. Is it possible to define independent measures (or combination of metrics) that can establish that the context is present (e.g. high BP and rapid pulse can indicate patient being in a state of high stress)

• 6. For the main potential effects listed below—can you define alerts that we may add to the guideline?

Main effects on: treatment goals, referrals, measurement type and schedule, drug type and schedule, exercise, diet routine, appointment schedule, daily routine (work, hobbies).

Thank you for your cooperation.

Appendix 4: Elicitation instrument for customizing clinical guidelines

The motivation of this questionnaire is to detect which psycho-social and demographic contexts affect treatment goals and plans. The questionnaire should be used with a specific clinical guideline considering the clinical decision points established in the guideline.

The process of elicitation of CIG-customized contexts (CCCs) is shown in Fig. 4 of the paper and is described here. In the first part of the elicitation process (top right of figure), knowledge engineers use general guiding principles for improving adherence and for reducing risks and complications, along with the psycho-social and demographic context-inducing concepts and general insights gleaned from the interviews that were previously conducted for eliciting these concepts and their effect types (as was described in part A of Section 3.2.2). Using these principles and insights, the knowledge engineers identify in the clinical guideline, and in its corresponding flowchart, sections in which risk, complications, and issues of non-adherence may arise. For example, adherence to short-acting insulin is difficult to achieve because the patient should have a routine daily schedule and diet; a non-routine schedule or diet increases the overall risk and the likelihood of complications. After the knowledge engineers focused on the sections that should be examined in the free-text clinical guidelines, in the second step (top left of the figure), expert clinicians define domain-specific scenarios that will focus on points of non-adherence to treatment, process risks, and process complications.

The next step is based on the domain-specific scenarios and on the psycho-social and demographic contexts identified by that point. In this step, the clinical experts define the domain-specific CCCs (e.g., semi-routine schedule) and their effects on treatment goals and recommendations (e.g., increase the number of daily measurements of blood glucose or blood pressure), which will enable the customization of the clinical guideline.

Psycho-social and demographic concepts describe different aspects of a patients’ condition and their environment, including family support, daily routine, stress, etc.

Each psycho-social and demographic concept is represented as a data item that has a range of values on a categorical scale. The range may depend on the clinical domain. A CCC may be defined as an expression over psycho-social and demographic concepts that limit the allowed range of the concept (e.g., relating to medium stress level). A patient may be in a context for a time interval.

Having a flowchart representing the structure of the guideline, we would like to go over it and check which psycho-social and demographic contexts could change a specific treatment step. Therefore, a first step would be to prepare this flowchart, focusing on decision points related to clinical state of the patient. Please number the decision points in this flowchart, so that we can refer to them during this interview.

We have identified, after more than 50 interviews with physicians and nurses, the following psycho-social and demographic concepts and their ordinal scales (a–e):

Psycho-social/demographic concept	a	b	c	d	e
Ability to comply with treatment	No ability	Minimal ability	Normal ability	Excellent ability
Communication level*	Low	Medium	High
Need for accompanying person for visits	Very high	High	Medium	Low
Support level	No support	Some support	Night support	Frequent support	Full support
Education level	Low	Medium	High
Daily/diet routine	No routine	Semi routine	Routine
Fitness level	Rest	Minimal	Normal	Moderate	Extreme
Language level	No knowledge	Basic	Fluent
Living area’s accessibility	Difficult	Moderate	Easy
Living area’s pollution level	Very high	High	Medium	Low
Distance from medical center	Isolated	Distant	Near by	Immediate
Stress status	High	Medium	Low
Financial capability	Low	Medium	High

* Communication level is an abstraction over cooperation level, desire to know truth about prognosis, education level, language level, and trust level

From our previous interviews, the treatment steps affected by the psycho-social and demographic contexts are usually of the following types:

• Medication change—changes in dose, schedule or type;

• Measurement modification—changes in schedule, type and amounts;

• Allowed concept threshold—changes in maximum or minimum values;

• Physical activity routine—changes in physical activity routine;

• Diet modification—changes to recommended diet;

• Appointment schedule—changes to timing of future appointments.

The questionnaire’s goal is to establish the treatments steps in the guideline that are affected based on the psycho-social and demographic context.

Psycho-social and demographic concepts	Affected scenario (considering context)	Effect on treatment or clinical goal
Ability to comply with treatment
Communication level*
Need for accompanying person for visits
Support level	If someone lives alone and had cerebral vascular accidents then starting insulin treatment may have a very high risk of hypoglycemia not being noticed	Then he doesn’t give insulin
Education level
Daily/diet routine
Fitness level
Language level
Living area’s accessibility
Living area’s pollution level
Distance from medical center
Stress status
Financial capability

• 1. Could you specify for the psycho-social and demographic concepts, which ones should be filled out by physicians (because the patient’s input might not be realistic)?

• 2. For the first part—looking at the guideline’s flowchart, please provide information regarding the following items:

  • Main clinical reasons that lead to non-compliance during treatments (For example, change in daily routine that leads to changes in meal composition or timing or even skipping a meal that will change the time of taking insulin in diabetes patients).

  • Think of clinical goals that are defined in the guideline, and identify which are the reasons that patient may not achieve these goals (For example, if asthma is not controlled using the current treatment regimen, treatment should be stepped up until control is achieved—which of the psycho-social and demographic contexts could be the reasons for not achieving controlled asthma state?).

  • Looking at the decision points, try to identify psycho-social and demographic contexts that would change the treatment step (For example, if there are several options to treat edema in a chronic patient and the best option requires family member support (concept), in case family members are not available (context), the physician may change the option to another option that the patient can handle in the new situation).

• 3. Please specify patients’ scenarios that you managed where his/her psycho-social or demographic context affected the treatment

• 4. Please use the following table to define for each psycho-social and demographic concept, the scenario that it affected and what the effect was.

Appendix 5: Pre-pilot testing

Before the evaluation study, we performed extensive testing during a pre-pilot period, to see that the system behaved correctly. Correct behavior implied correct and timely response to user input, to context switches, and to patterns discovered in the data. It also meant that the BEE-DSS changed plans and projected new knowledge to the mDSS. We tested that the mDSS and BE-DSS delivered correct and timely recommendations to care provider and patients. We further tested that the mDSS delivered reminders to patients according to current patient model (that considers current patient clinical data and current therapy), while addressing timing of reminders and active context. The testing for the AF domain was done by nine AF and seven GDM patients who used the system for 2 months. They asked the volunteers to test all monitoring and reminders, and the “pill in the pocket” evidence-based protocol, that for eligible patients (depending on their medication therapy and clinical and social parameters) provided indication on when to take an emergency drug that the patient was instructed to keep in his pocket and take when an AF event was experienced. The protocol involved responding to patients reports of unacceptable symptoms, sending them a recommendation to conduct a 30-min ECG monitoring. If an AF episode was detected in high certainty and the patient has not taken the emergency drug in the past 4 h, s/he was sent a recommendation to take the emergency pill and then test the ECG after 1 h.

The AF clinicians tested the care provider’s parallel workflow consistently uses (like the patient workflow) the patient model and provides appropriate recommendations. In addition, the clinicians tested the cardioversion advice protocol.

To test the GDM system during the pre-pilot stage, we developed a simulation program that generated patient data for long periods of time because the temporal patterns that were monitored spanned several weeks (e.g., 30 days of good blood glucose values). The aim of the simulation program was to test all the different workflows covered by the GDM guideline. Four clinicians and three knowledge engineers involved in the guideline elicitation tested the system for 2 months in a daily basis, in terms of EMR data imports with simulated data, entry of GDM relevant monitoring parameters, checking of the content of technical and clinical recommendations and reminders generated both at patient and caregiver GUIs. Monitoring data entered by users during this period included simulated glucose values, diet, ketonuria, insulin, measurement of blood pressure with a sphygmomanometer and real physical activity sessions performed with a Physical Activity Detector (PAD) integrated in the mobile application. Bugs in software components and in the knowledge base were fixed for both domains.