Using Knowledge Modelling to Measure How Clinical Practice could - - PowerPoint PPT Presentation

Context Material Method Results Discussion and conclusion

Using Knowledge Modelling to Measure How Clinical Practice could Actually Be Evidence-Based:

a Preliminary Analysis with Arterial Hypertension Management Brigitte S´ eroussia, Jacques Bouaudb, Denk´ e L. Denk´ ea, Hector Falcoffc Jacques Juliend,

aAP-HP, Tenon, D´

epartement de Sant´ e Publique & UFR de M´ edecine, Universit´ e Paris 6 & APREC, Paris, France & LIM&Bio, Bobigny, France

bAP-HP, DSI, STIM & INSERM, UMR S 872, eq. 20, Paris, France c SFTG, Paris, France; Universit´

e Paris 5, Facult´ e de M´ edecine, D´ epartement de M´ edecine G´ en´ erale, Paris, France.

d AP-HP, HEGP, Service de M´

edecine Vasculaire et Hypertension Art´ erielle, Paris, France. S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 1/20

Context Material Method Results Discussion and conclusion

Medical practice and medical knowledge

Continuous publication of medical knowledge Observed variability of medical practice

Differences between state of the art and practice Suboptimal healthcare quality - Medical errors

The paradigm of evidence-based medicine

“[...] the explicit, judicious, and conscientious use of current best evidence from health care research in decisions about the care of individuals and populations.” 1

1[Sackett et al. Evidence based medicine: what it is and what it isn’t. Br Med J 1996;312(7023):71-2] S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 2/20

Context Material Method Results Discussion and conclusion

Clinical practice guidelines

Clinical practice guidelines (CPGs)

Promotion of professional best practices in health care Health professional societies or national agencies Systematic reviews of publications Publication as textual documents “catalog” of recommendations: prototypical situations with corresponding recommended actions

Recommendation strength

Recommendation with grade: A, B, or C ➠ supported by scientific evidence Recommendation without explicit grade ➠ supported by consensus within CPG working groups

Implementing guidelines: so what?

Almost no impact of textual dissemination Existence of knowledge “gaps”

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 3/20

Context Material Method Results Discussion and conclusion

Guideline-based clinical decision support systems

Guideline-based clinical decision support systems (CDSSs)

CPGs are the knowledge resource Computerized model of CPG content Patient-centered recommendations at decision time (Some) evidence of impact on practice

Modeling the knowledge base of a CDSS

Knowledge engineering approach From explicit textual statements to a formalized representation Often requires interpretation assumptions

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 4/20

Context Material Method Results Discussion and conclusion

The guiding mode of ASTI (ASTI-GM)

ASTI project (Computerized Support for Therapeutic Strategy)

A prototype guideline-based CDSS Therapeutic prescribing in primary care AHT, dyslipidemia, diabetes, smoking cessation, AF Patient follow up, order critiquing, guiding module

ASTI guiding mode

Knowledge base as a decision tree A nosological repository of patient profiles with associated therapeutic recommendations Systematic modelling

Exhaustive coverage through a completion step Combination of different knowledge sources Synthesis and conflict resolution with an expert

➠ Output might not be explicitly supported by CPGs

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 5/20

Context Material Method Results Discussion and conclusion

Problem statement and objectives

What is the evidence status of the recommendations issued by ASTI-GM?

Which recommendations correspond to evidence-based statements, to consensus-based statements in the CPGs, or are not explicitly supported by the CPGs? Which ones are used in practice? How clinical practice could actually be evidence-based?

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 6/20

Context Material Method Results Discussion and conclusion

Problem statement and objectives

What is the evidence status of the recommendations issued by ASTI-GM?

Which recommendations correspond to evidence-based statements, to consensus-based statements in the CPGs, or are not explicitly supported by the CPGs? Which ones are used in practice? How clinical practice could actually be evidence-based?

Objectives

1 Characterize the evidence status of a recommendation in

ASTI-GM KB

2 Analyze the status theoretical distribution in the KB 3 Analyze the status distribution on a patient sample S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 6/20

Context Material Method Results Discussion and conclusion

Modelling AHT management

French AHT management CPGs (2005) AHT management principles

Identification of the clinical situation (comorbidities, risk factors) Identification of the current treatment wrt recommended treatment plan [T1, T1+T2, T1+T2+T3] ➠ In case of inefficiency, add drug ➠ In case of contraindication, substitute drug

Modelling principes2

Atomisation, deabstraction, and desambiguisation steps Knowledge completion step:

Implicit criteria Implicit situations Contribution of implicit medical knowledge

2Shiffman et al. Bridging the guideline implementation gap: a systematic, document-centered approach to guideline implementation. J Am Med Inform Assoc 2004, 11(5);418-426. S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 7/20

Decision tree

Coronary disease Antecedent of stroke Renal disease Diabetes Microalbuminury Ederly Current treatment Treatement type CI to ACEi yes no yes no no yes no yes no yes yes no yes no mono bi tri quadri and more Prior tt by ARBs + TDs Efficient treatment yes yes no no

Clinical level Therapeutic level

no

ARBs + TDs + CCBs + statin + aspirin + HM ATCD of stroke diabetes elderly bitherapy CI to ACEi prior tt by ARBs + TDs ARBs + TDs + CCBs + statin + aspirin + Health meas.

Non explicitly supported recommendation

Context Material Method Results Discussion and conclusion

Data collections

1 Extended KB of ASTI-GM

Path in the decision tree = a conjunction of decision criteria + therapeutic recommendations ➠ 44,571 theoretical patient profiles

2 A data sample of actual hypertensive patient

Extracted from a GP medical practice (March 2007) Selection of hypertensive patients Selection of orders posterior to December 2005 with antihypertensive drugs Collection of patient data at decision date ➠ 435 actual patient profiles

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 10/20

Context Material Method Results Discussion and conclusion

Evidence status definition

Assumption: The status of a recommendation can be characterized from the criteria of its patient profile Definition: A profile P has the evidence status of the recommendation it supports, or is eligible for 3 evidence status:

Evidence-Based (EB): explicit statement with grade (A, B, C) Consensus-Based (CB): explicit statement without grade Non-Supported (NS): no explicit statement in the CPGs

3 variable categories reusing ASTI-GM modelling

Clinical criteria: 9 comorbities (elderly, antecedent of stroke, myocardial infarction, stable angina, ischemic cardiac failure, diabetes, microalbuminury, renal disease, left ventricle hypertrophy) Current level of drug association (no treatment, mono-, bi-, tri-therapy...) Drug contraindications (CI to ACEi or BB)

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 11/20

Context Material Method Results Discussion and conclusion

EB profiles

P is “clinically EB” (EBcl) if there exists at most one comorbidity among elderly, antecedent of stroke, MI, stable angina, ischemic cardiac failure, (diabetes ∧ microalbuminury) for which explicit EBM statements are provided in CPGs. P is “therapeutically EB” (EBth) when treatment is initialized (no current treatment), for which evidence exists. P is “EB with respect to contraindications” (EBci) if there are no contraindications (except in the case of ischemic cardiac failure where a graded recommendation exists when ACEi are contraindicated). EB(P) = EBcl(P) ∧ EBth(P) ∧ EBci(P)

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 12/20

Context Material Method Results Discussion and conclusion

CB profiles

P is “clinically CB” (CBcl) in some contexts explicitly described in CPGs and for which consensus-based advices are provided: renal disease alone, left ventricle hypertrophy alone, (renal disease ∧ diabetes), (myocardial infarction ∧ ischemic cardiac failure). P is “therapeutically CB” (CBth) when the current treatment level is at most a bitherapy since these CPGs recommend up to tritherapies. P is “CB with respect to contraindications” (CBci) only for CI to ACEi (except when explicitly mentionned with grade) CB(P) = ¬EB(P) ∧(CBcl(P) ∨ EBcl(P)) ∧(CBth(P) ∨ EBth(P)) ∧(CBci(P) ∨ EBci(P))

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 13/20

Context Material Method Results Discussion and conclusion

NS profiles

P is NS, when it is neither EB nor CB. For these profiles, there is no explicit recommended action in CPGs: NS(P) = ¬EB(P) ∧ ¬CB(P)

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 14/20

Context Material Method Results Discussion and conclusion

Implementation and experiments

Implementation of a status function from a patient profile Application to the KB theoretical profiles Application to the actual patient profiles

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 15/20

Context Material Method Results Discussion and conclusion

Distribution of theoretical patient profiles

Theoretical distribution (n = 44, 571) Profile status EB CB NS Total n 206 5,424 38,941 44,571 12.6% of the KB account for explicit CPG statements

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 16/20

Context Material Method Results Discussion and conclusion

Distribution of actual patient profiles

Actual distribution (n = 435) Profile status EB CB NS Total n 36 204 195 435 91.7% of the patients are not supported by EB statements Almost half (44.8%) are “out” of CPGs

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 17/20

Context Material Method Results Discussion and conclusion

Discussion and conclusion

Modelling aspects in ASTI-GM

Extension of the theoretical coverage by completion Many non CPG supported recommendations (“Expert-based”) EBM proportion in the KB is weak

In the data set, EBM effective in less than 10%

EB statements apply only once

At most one referenced comorbidity First antihypertensive order No contraindication

Difficulties in the management of drug sequences

Characteristic of chronic disease management Supported by CB statements in 47%

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 18/20

Context Material Method Results Discussion and conclusion

Discussion and conclusion - continued

Importance of NS patients (44,8 %)

Limited help for GPs from CPGs Explanation for practice variability? Which role CDSSs can play for NS patients?

Effect of profile status on ASTI-GM compliance

Relationship between ASTI compliance and evidence status? Nonsense of CPG compliance for NS profiles

Limits of a case study

AHT wrt French CPGs Non representative sample

Perspectives and questions

Same results for other patient samples, other CPGs, other pathologies? Qualify ASTI recommendations with their status

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 19/20

Context Material Method Results Discussion and conclusion

Thank you

S´ eroussi, Bouaud, Denk´ e, Julien, Falcoff MIE 2009, Sarajevo, BH 20/20