Using Scenarios in Chronic Disease Management Guidelines for Primary Care

Using Scenarios in Chronic Disease Management Guidelines for Primary Care Peter D. Johnson MB BS1, Samson Tu M.S.2, Nick Booth MA MB BS MRCGP DCH 1, Bob Sugden MBCS 1, Ian N. Purves MB BS, MRCGP, MD 1

1Sowerby Centre for Health Informatics in Newcastle

University of Newcastle upon Tyne, Newcastle Upon Tyne, UK

2Stanford Medical Informatics

Stanford University School of Medicine, California, USA

The P RODIGY system is a guideline-based decision-support system designed to assist general practitioners in England choose the appropriate therapeutic action for their patients. As part of the system, we developed a novel model for encoding clinical guidelines for managing patients with chronic diseases such as asthma and hypertension. The model structures a guideline as a set of choices to be made by the clinician. It models patient scenarios which drive decision making and are used to synchronize the management of a patient with guideline recommendations. The model is robust with respect to available input data and leaves the control of decision-making to the clinician. We have built execution engines to verify the computability of the model. We intend to test the model integrated in up to 200 live systems from at least four system vendors in English General practice.

INTRODUCTION

In the United Kingdom, the National Health Service (NHS) has funded the P RODIGY* Project to develop a guideline-based decision-support system that can assist general practitioners (GPs) in the task of choosing rational therapeutic actions for their patients. The P RODIGY system has gone through two previous phases of development and evaluation. In each phase, the P RODIGY team created specifications for the behavior of the software system and created computer representation of clinical guidelines taken from literature. P RODIGY Phase I and II systems were implemented by system vendors as modules extending their existing Electronic Patient Record (EPR) systems, using a similar look-and-feel, and accessing the coded information in the EPR to help direct choices in the guidelines. There is one guideline per diagnosis, with possible patient situations within that diagnosis organised into scenarios. Choice of a scenario results in a list of possible actions *P RODIGY stands for (P rescribing R ati O nally with D ecision-support I n G eneral-practice stud Y)conditioned on findings in the EPR. Actions include on-line explanations, printed drug prescriptions and printed patient information leaflets. The release version of P RODIGY Phase II is now rolled out to most GPs in England.

P RODIGY II has been shown to be technically competent at acute disease management for advice, shared decision making and prescribing. However, the representation of guidelines for chronic diseases in P RODIGY II was judged by clinicians to be inadequate; users followed P RODIGY advice for chronic disease one third as often as acute disease1. In addition guideline authors found creating guidelines for the management of chronic disease in P RODIGY a difficult task with unsatisfactory results.

Phase III of P RODIGY was initiated to develop a system of guidance for chronic disease management in primary care, using a model that the guideline authors and general practitioners found intuitive. A primary aim was that users must be able to match patients to the guideline when they started part way through the guideline’s flow.

Characteristics of Chronic Disease Management Asthma, hypertension2 and angina are examples of chronic disease for which evidence-based guidelines exist for primary care. The main difference between management of acute and chronic disease is the consideration of time. Decisions about the care process are dependent on decisions made and actions taken at previous consultations, and the outcomes of those actions. Notions of the state of disease control over time are important (‘controlled’, ‘uncontrolled’or ‘critical’) as is the trend (‘worsening’ or ‘improving’). The concept of a therapy that persists over time but can be modified is important, e.g. taking ‘inhaled steroids’, modifiable by the separate attributes of dose quantity, dose frequency and delivery device. Thus a patient whose asthma remains uncontrolled on a short-acting beta2 agonist and low-dose inhaled steroid, may either ‘step up’ the dose of inhaled steroid to medium-dose, or alternately start a long-acting beta 2 agonist at a standard dose level.

METHOD

Design principles were stated as follows:

?The guideline model must cope with all chronic disease management guidelines for primary care.?The system must minimise user interaction as consultations are of short duration.

?The clinician must always be able to make his/her own choice of actions, overriding any system suggestions. The rationale for this is that no guideline can cater for every combination of circumstance, and the EPR is unlikely to contain all the information necessary to make a completely informed decision.

?The system should predict what scenario is most likely at any given consultation, using information stored from the last consultation and data entered in the EPR. However, the user must always have the ability to override and reposition the patient in the guideline.

?The user should be able to ask ‘what if’ questions of the system – “if I take this decision, what does it result in me doing?”

?The system should suggest clinician data entry where the data supplied would be useful in refining choices of scenarios or actions, (shortening lists of choices) but that this data entry should always be optional.

We observed a tension between producing a guideline representation that is expressive, and one that the guideline authors and clinicians can understand and utilise productively. We aimed to produce the simplest guideline model sufficiently expressive to represent chronic disease management guidelines in a general practice setting.

We made use of Protégé3, a knowledge modelling environment developed at Stanford University, to test rapidly whether the model we created was usable and expressive enough for the guideline authors. Guidelines were written for the management of asthma, hypertension and stable angina. A prototype Execution module was built as a test of the model’s computability. Regular meetings have been held with system vendors to demonstrate the guidelines and prototypes to gain feedback. As a consequence several refinements to the design of the execution module were made, with final implementation as an ActiveX control. The system vendors are using this in their systems to provide the guideline model behaviour. The vendors provide a data interface to their EPR and a user interface that integrates with the look and feel of their existing system.

Workshops with potential users demonstrated the critical nature of an intuitive user interface, and the need for a simple user interaction model, in turn influencing the guideline model.

GUIDELINE MODEL

The P RODIGY III model divides a chronic disease management guideline into two sections. The first describes the management sequence over time, demonstrating the therapeutic choices available for each recognisable state of the disease being managed. This offers a strategic view – a map of the whole area of disease management covered by the guideline. A patient should ideally become stabilised in a particular area of this map, moving only as the disease or management of the disease changes. Information in the EPR is used to suggest which management choices are preferred and which are ruled out. The second section (the consultation template) describes actions that may be taken each time the patient is seen by a clinician. This section may be specialised for each scenario, and contains actions which are either present to inform the decision making process (data collection, investigations) or are opportunistic in nature (e.g. education).

Scenarios, Action Steps, and Subguidelines

We represent a high level view of a guideline as a network consisting of scenarios, action steps, and subguidelines. A layer of detail below this navigation layer provides refinement of decisions and recommendations using criteria that operate using information in the EPR to express degrees of preference for each of many possible actions. Experience from earlier P RODIGY phases suggests that the concept of ‘scenario’ is useful to both GPs and guideline authors. A scenario is an easily recognisable patient state for a particular diagnosis e.g. ‘angina on triple therapy’, or ‘hypertensive on non-pharmacological treatment’. These easily recognisable states are usually, but not always, a combination of diagnosis and current therapy.

As many patients join a guideline part way through treatment for their disease, or have part of their treatment at other sites, there is an overriding need for easy positioning of the patient within the guideline. As the EPR is rarely comprehensive or completely accurate (missing or incorrect data), one cannot rely on the system to always position the patient in the guideline. Scenarios provide an intuitive set of access points into the guideline for the clinician when the EPR does not reflect the true management state. Scenarios are not provided for all possible states in the management of a disease, as this would lead to a combinatorial explosion of states. Instead a few high level states are used as access points, refinement of the

state being carried out successively in action steps and subguidelines.

For each scenario there are a number of outcome assessments , and one or more action steps may be suggested to manage each outcome. A ‘hypertensive on monotherapy’ may be assessed as ‘BP > 160/100’with possible action steps of ‘increasing dose’ or ‘add second agent’. The action to increase dose will work out which agent the patient is on in a subguideline,and only be possible if the patient is not already on the maximum dose for this agent. The action steps can initiate, modify or stop activities that persist over time, (e.g. betablocker therapy) or execute instantaneous actions such as printing out a patient information leaflet.

An action step is a potential choice of management for a given scenario and outcome assessment. It is used in simple management decisions. For a scenario of ‘BP last visit < 150/90’ in the angina guideline, one of the action steps has an assessment label of ‘B P today < 150/90’ and suggested actions of ‘schedule appointment for 6 months’. These actions may be recommended or ruled out by criteria, which are discussed further below. In the case where no criteria can be evaluated, either because none has been written or because of lack of necessary data, a default preference can be used. An action step may also contain references to evidence; these are expressed as one or more URLs.

In certain cases a single action step is not sufficient to describe the intended process of care, and further decisions and rules are necessary to refine the action steps chosen. In these cases a Subguideline step is used. A subguideline describes a further set of scenarios and action steps to be assessed within the

consultation.

For any given scenario there may be an attached Consultation template that specifies best-practice actions, usually to gather more data relevant to the current therapeutic decision. This data is then used by criteria to rule-out or express preference for action steps. Other opportunistic data collection and education actions may reside in this section also. The consultation template is optional for the clinician.

Two aids are used by the system in working out which scenario is most likely. Firstly, for a patient who is already ‘on-guideline’ the action taken last is used to suggest the most likely scenario at this visit. Secondly,each scenario has a precondition (criterion) that must be true for this scenario to be possible. For example for the ‘hypertensive on dual therapy’ scenario to be possible, the patient must have current EPR entries for two antihypertensive drugs. In cases where the current information in the EPR suggests the predicted scenario is wrong, preconditions are used to suggest the subset of most likely scenarios in the guideline.For any one scenario, there may be many possible action steps. Criteria are evaluated using the EPR and data entered as part of the consultation template to rule out unsuitable action steps and to express which action steps are preferred.

In cases where there is no relevant information in the EPR, and where the user enters no data in the consultation template, the system will be unable to rule out, or express preference for any action step. In this case the user will get a complete list of possibilities, from which they must choose manually.Thus the model copes with insufficient information gracefully, and performs better the more information present in the EPR or entered during the consultation.

Figure 1 High level view of Hypertension guideline, with six top level scenarios

Actions and Activities

In the P RODIGY III model, clinical interventions have been split into Actions, which are effectively instantaneous as far as primary care is concerned, and Activities, interventions that are started and persist until they are modified or stopped. As the project is primarily concerned with drug prescribing, the activity on which we focused our modeling effort is the treatment regimen, a high-level abstraction of a class of drug therapy, such as ‘inhaled steroid therapy’ in an asthma guideline. Regimens are acted on by state-changing Actions. Thus starting a patient on ‘low-dose inhaled steroids’ involves a start regimen action, which acts on the inhaled steroids regimen. Actions are divided into ones that act on regimens:

?Start / Stop regimen

?Step up / Step down dose

?Step up / Step down frequency

?Change preparation

and other instantaneous actions:

?Acute prescription

?Print patient information leaflet

?Data entry

?Order investigations

?Referral

?Schedule follow up appointment

?Display information at a URL

Regimens and Regimen Components.

A drug regimen such as ‘inhaled steroids’ contains a number of regimen components. A component may be uniquely specified for dose, frequency and preparation type. Thus the inhaled steroid regimen contains 12 components to cover the three axes of variation e.g. a ‘low-dose MDI’ (Multi-dose inhaler) component, and a ‘Medium-dose, dry powder’ (higher dose, different preparation). Inhaled steroids for asthma is the only regimen we have currently encountered which requires use of all three axes of modification. The thiazides regimen in hypertension only varies along the dose axis.

Each component may contain a number of alternative prescribable items, i.e. specific prescriptions where the dose, frequency and preparation are equivalent. Each prescribable item contains all the information for printing a drug prescription — the drug, dose, frequency, preparation and quantity. The low-dose MDI component of the inhaled steroid regimen, for example, contains the following prescribable items:?Beclomethasone MDI 100mcg twice daily, 1 pack ?Budesonide MDI 100mcg twice daily 1 pack Regimen actions such as ‘step up dose’ when applied to the inhaled steroid regimen may change a patient from taking a low-dose MDI containing Beclomethasone to a medium-dose MDI component, containing a prescribable item of the same drug but at 200mcg twice daily.

Criteria

The preconditions of scenarios, the rule out and rule-in preferences of action steps and the preference criteria of prescribable items are implemented as Boolean expressions to express a preference for or against a choice. The criteria may refer to the presence of NHS Clinical Terms in the EPR within a time period, e.g. ‘absence of cough within last 3 months’, or use numeric values held with coded terms in the EPR, e.g. ‘systolic blood pressure > 180mmHg within the last year’. Atomic criteria may be combined using Boolean combinations (AND, OR, XOR).

Drug related criteria include ‘presence of sensitivity to penicillin’, ‘presence of inhaled steroid therapy within last month’, ‘presence of two antihypertensive agents currently’. Much of the knowledge required to analyse whether medication entries in the EPR satisfy these criteria is generally true about drugs, and is not specific to any one guideline. A separate drug knowledge base project4 has been undertaken as part of P RODIGY to supply this knowledge.

Abstraction

Guideline authors intuitively express criteria in terms of abstract concepts such as ‘cardiovascular risk’. This requires the system to infer new information from existing information in the EPR. To enable this, we provide a set of functions with which abstract concepts, such as the risk rating, can be computed from more elementary concepts such as BP, smoking, cholesterol level, and other risk factors.

INTERACTING WITH A GUIDELINE For a new patient with a chronic disease not already managed on guideline, the P RODIGY III system will suggest a list of scenarios that may be appropriate. This is achieved by evaluating the preconditions for all scenarios in the guideline, and displaying a subset of the list from which the clinician may choose. For a patient who is already on guideline, the system first suggests the most likely scenario, based on actions taken at the last visit. If in the hypertension guideline at the last visit a low-dose betablocker was started, namely ‘Atenolol 25 mg tabs once daily’, the first scenario suggested at this visit would be ‘hypertension on monotherapy’. If this is no longer correct, (e.g. the patient had their medication changed elsewhere) the system will evaluate the preconditions of all scenarios and provide an appropriate subset in the same way as starting a guideline afresh.

Once a scenario is chosen, a list of possible

assessments with proposed actions is displayed. This list may have some proposals ruled out and some preferred, as a result of the criteria acting on information entered via the consultation template, or from existing information in the EPR. Thus with the ‘hypertension on monotherapy’ there are 5 possible choices. Each assessment may have suggested actions — for ‘BP > 160/90’ two alternatives are given, increase current therapy (if not already on maximum dose), or start a second agent. If the user chooses to increase the current therapy, as they are on Atenolol 25mg, the system marks Atenolol 50mg as the preferred choice. Accepting this recommendation results in the drug prescription being printed and the resulting changes made to the EPR, including storage of the position in the guideline for use at the next visit. With a few clicks – as few as four, the clinician benefits from guideline assisted treatment choice and good practice data entry, without artificial constraint.

DISCUSSION

Many guideline models exist which aim to encapsulate interactive guidelines in conjunction with the EPR. The simplest are flow diagram like models, where the user is rigidly guided down a path with an enforced, arbitrary sequence of decisions and actions. Earlier P RODIGY phases fall into this category. This approach does not cope well either with care over many visits or with the way GPs think about chronic disease. While it is easy to create simple guidelines using flow diagrams, scaling up to the complexity of chronic disease care produces unwieldy systems.

The P RODIGY III guideline model differs from GLIF5, EON6, and Asbru7 in several ways. It aims to provide guidance on choices available in any situation instead of recommendations based on prescriptive plans of actions. Act management has been simplified from these models that can provide more rigorous synchronisation of parallel actions with duration, as rarely is this relevant to primary care. Instead of requiring guidelines to recognize or infer complex patient conditions, such as temporal abstractions used in EON and Asbru, (which require good EPR data) the emphasis is on interactive use where a clinician can be relied on to recognize complex patterns.

P RODIGY III differs also in utilising Scenarios to provide the key entry points to the guideline and make it easier to synchronise guidelines to patients already on treatment. The limitations of a pure state-transition model do not arise because of the ability to refine choices in actions steps and subguidelines.

A clear distinction is made in the P RODIGY III guideline model between disease management actions, and optional actions that generate more information.Systems built using the model can express more refined preferences when more data is available. A deliberate effort was made to simplify the model as far as possible, to enable the creation of many guidelines and to facilitate their implementation in live primary care systems.

Protégé was invaluable in the rapid development of guideline models and in testing the usability of those models by the authors. Dev elopment of execution modules and evaluation with guidelines closed the design loop, enabling us to have confidence in the system and verify the guideline content.

At the time of writing the execution module is being incorporated in several vendor systems for a formal evaluation in primary care over the next year.

Acknowledgements

Funding provided by NHS Executive Primary Care and project management by Mike Sowerby. Thanks to SMI in enabling Samson Tu’s sabbatical at Newcastle.

References

1.Report on the Results of PRODIGY Phase Two,

NHS Executive (Primary Care Branch) Available from URL:

https://www.wendangku.net/doc/757654770.html,/prodigy/reports/P2SUMM-final.pdf 2. Ramsay, L.E. and et.al. (1999) Guidelines for

management of hypertension: Report of the third working party of the British Hypertension Society.

Journal of Human Hypertension 13, 569-592.

3. Musen, M.A., Gennari, J.H., Eriksson, H., Tu,

S.W., Puerta, A. R. PROTéGé-II: Computer Support For Development Of Intelligent Systems From Libraries of Components. Proceedings of MEDINFO '95, Vancouver BC. 1995; 766-770. 4.Solomon WD, Wroe CJ, Rector AL et al. A

Reference Terminology for Drugs. Journal of the American Medical Informatics Association 1999;

Fall Symposium Special Issue.

5.Ohno-Machado L, Gennari JH, Murphy S, et al.

The GuideLine Interchange Format: A Model for Representing Guidelines. Journal of the American Medical Informatics Association 1998;5(4):357-372.

6.Musen M, Tu S, Das A, Shahar Y. EON: A

Component-Based Approach to Automation of Protocol-Directed Therapy. JAMIA 1996;3:367-388

7.Shahar,Y Miksch, S & Johnson, P. An Intention-

Based Language for Representing Clinical Guidelines. In James J. Cimino, Ed., 1996 AMIA Annual Fall Symposium, Washington, D.C., 592-596. Hanley & Belfus, 1996.