臨床決策支援系統

(1)

輔仁大學

醫學資訊與創新應用學士學位學程

醫療標準及術語

臨床決策支援系統

Clinical Decision Support System

臺北市立聯合醫院仁愛院區家庭醫學科郭冠良

Kuan-Liang Kuo, M.D., Ph.D.

2022-05-24

1

(2)

Topics

• Decision Making in Healthcare

• Clinical Decision Support System

• Knowledge Representation

• CDSS Implementation

• Evaluation of CDSS

(3)

Decision Making in Healthcare

Reference:

Gotzsche, Peter.

Rational Diagnosis and Treatment: Evidence-Based Clinical Decision-Making. 4th Ed. Willey.

(4)

Introduction

(5)

Clinical Decision Making

• This complex task needs:

– Knowledgeable practitioner – Reliable informational inputs – Supportive environment

(6)

The Foundation of Clinical

Decisions

(7)

Clinical Decision Process

Gotzsche, Peter.

Rational Diagnosis and Treatment:

Evidence-Based Clinical Decision-

Making. 4th Ed. Willey.

(8)

Clinical Data

• 與臨床決策有關的病人所有資料

• Clinical data  Clinical picture

• Types

– Subjective – Objective

• 理學檢查

• 實驗室檢查

(9)

Scales of Measurement

• Nominal Scale

– 明確定義的類別 (well-defined) – 隸屬於單一類別 (exclusive)

– 類別必須含括所有事件 (exhaustive) – 例:

• 頭痛

• 發熱

• 大腸息肉

• Ordinal Scale

• Interval Scale

(10)

Scales of Measurement

• Nominal Scale

• Ordinal Scale

– 例:

• Beaufort scale of wind force

• Muscle Power

• 脂肪肝

• Interval Scale

0-- 無肌肉收縮

1 -- 些微肌肉收縮，但無移動

2 -- 無重力影響下可自由活動 (如放在床上水平移動)

3 -- 可對抗重力 (如可將手自下垂狀態提起) 4 -- 可對抗重力且能部份抵抗外力

5 -- 可完全抵抗外力而無困難(為正常狀況)

(11)

Scales of Measurement

• Nominal Scale

• Ordinal Scale

• Interval Scale

– Continuous

– Discontinuous (Discrete) – 例

• 體重

• 體溫

– May be reduced to ordinal or binary scale

(12)

Taking the History

• Symptoms

• Personal history

• Past history

• Family history

• Others

• 望聞問切

(13)

The Physical Examination

• Head and neck

• Chest and heart

• Abdomen

• Extremities

• Skin

• Neurology

• Others

(14)

The Physical Examination

• Terminology

– Chaotic – 例

• Breathing sound

– Coarse – Wheezing – Rales

– Friction rubs

– Computer-aided education

• Typical cases

(15)

Paraclinical Findings

• Laboratory

• Radiology

– 描述 vs 診斷

• Pathology

– Gold standard – 結果 vs 過程 – 局部 vs 全面

(16)

Further Examinations

• Ultrasonography

• Endoscopy

• 肌電圖，神經傳導

• 心電圖

• 肺功能

(17)

Global Assessments

• General appearance

– ill-looking / well-looking – Clinical Index

• 例: Heart function classification by NYHA

– Class I: 活動無限制

– Class II: 正常的活動就會累、心悸、或喘

– Class III: 小於正常的活動就會累、心悸、或喘 – Class IV: 休息時就會有症狀

(18)

Reliability and Relevance of

Clinical Data

(19)

Clinical Data on an Interval Scale

• 多次重複分析同一檢體

(20)

Clinical Data on an Interval Scale

(21)

Clinical Data on an Ordinal Scale

• 症狀嚴重度

• 疾病嚴重度

• 缺點

– 可再製性低 (reproducibility) – 不精確

– 須與病人實際狀況關聯

(22)

Clinical Data on an Binary Scale

• Descriptive paraclinical findings

– Chest X-ray (CXR)

• Small chest film  Positive 

• Large chest film

– Inter-observer agreement

• 不同醫師

– Intra-observer study

• 同一醫師，看兩次

(23)

Clinical Data on an Binary Scale

– Observed agreement

• (11 + 35)/60 = 77%  p0

– Expected rate of chance agreement

• 兩次都判讀為陽性之個案數

– 22/60 * 14/60 * 60=5.13

• 兩次都判讀為陰性之個案數

– 38/60 * 46/60 * 60=29.13

• 兩次都判讀相同結果之機率

– (5.13 + 29.13)/60=57%  pc

– Kappa statistics

• Observed agreement與 Change agreement的相關

• (p0 – pc) / (1 – pc)

• = (77%-57%)/(1-57%)

• = 0.47

(24)

Clinical Data on an Binary Scale

– Kappa statistics

• = 0.47

– Kappa = 1

• Perfect agreement

– Kappa = 0

• Chance agreement

– Kappa < 0

• Systemic disagreement between the observers

(25)

Clinical Data on an Binary Scale

• Descriptive physical signs

– Expected kappa: mean of values guessed by doctors

(26)

Clinical Data on an Binary Scale

• Descriptive physical signs

– Reasons for the low reproducibility of descriptive data

• Experience

• Thoroughness

– Repeat studies

– Standardize the descriptions

(27)

The Statistical Concept of Normality

• Define normal range of a laboratory test

– 95% of the healthy population

• 1/20 false positive

– No prior assumption for the shape of the distribution

(28)

The Statistical Concept of Normality

– 2.5 ~ 97.5 percentiles – Mean ± 2 SD

• Gaussian distribution

(29)

The Statistical Concept of Normality

– Possible reference population problem

• Age

• Sex

• 潛在疾病

• …

(30)

The Statistical Concept of Normality

– Question:

• 一個健康人接受20次檢查，有一次異常報告的機會?

– 5%

– 14%

– 36%

– 64%

– 72%

(31)

The Statistical Concept of Normality

– Question:

• 一個健康人接受20次檢查，有一次異常報告的機會?

• 1-0.95²⁰=0.64=64%

(32)

The Concept of Disease

• Clinical

– Normal – Abnormal

• Statistical

– Normal distribution

(33)

The Concept of Disease

• Somatization

• Bio-psycho-social disease concept

• Health check-up

• Mass screening

• Drug industry

– Inventing new diseases – Promoting diseases

(34)

Interpretation and Relevance

• Regression towards the mean

– Lab test

• 1^st time: Abnormal

• 2^nd time: Normal

– 治療效果

• 感冒

(35)

Indicators

• Clinical data as indicators

• 目的

– 診斷 (diagnosis) – 預後 (prognosis) – 嚴重度

• 種類

– Scale – Ordinal – Binary

• 資料來源

– 主訴

– 理學檢查 – 檢查、檢驗

(36)

The Disease Classification

(37)

The Theoretical Perspective

• The mechanical model of disease

– Structure – Organism

– Physiological mechanism change

• Complex feedback mechanisms

• Self-regulation of the system

– Re-establish balance

• Molecular biology

– DNA

• HBV

– Protein

(38)

The Theoretical Perspective

• Disease and the environment

– Population density and infectious diseases

• Peptic ulcer

– Helicobactor pylori

• AIDS

(39)

The Theoretical Perspective

• Classical

– Linear dynamics

• Modern

– Nonlinear

– Complex feedback mechanism – Chaos theory

• 非線性系統具有的多樣性和多尺度性。混沌理論解釋了決定系統可能產生隨機結果

(40)

The Practical Perspective

• Disease

– Definition – Cause

• Etiology

• Pathogenesis

– Clinical Picture – Diagnosis

– Treatment – Prognosis

(41)

Causes of Disease

(42)

Diagnosis

(43)

The Diagnostic Universe

(44)

The Diagnostic Universe

• Diagnostic probabilities

• P(D+|T+)

– 有病/陽性

• P(D-|T+)

– 無病/陽性

• P(D+|T-)

– 有病/陰性

• P(D-|T-)

– 無病/陰性

(45)

The Diagnostic Universe

• Nosographic probabilities (病情學)

• P(T+|D+)

– 陽性/有病

• P(T-|D+)

– 陰性/有病

• P(T+|D-)

– 陽性/無病

• P(T-|D-)

– 陰性/無病

(46)

Diagnosis of diseases with an

Accessible Defining Criterion

(47)

Diagnosis of diseases with an Accessible Defining Criterion

• Bayes’ theorem

(48)

Diagnosis of diseases with an Accessible Defining Criterion

• Specificity

– P(T-|D-)

• Sensitivity

– P(T+|D+)

• Predictive value of positive test

– P(D+|T+)

• Predictive value of negative test

– P(D-|T-)

(49)

Diagnosis of diseases with an

Accessible Defining Criterion

(50)

Diagnosis of diseases with an

Accessible Defining Criterion

(51)

Diagnosis of Diseases with a Concealed Defining Criterion

• 診斷條件不明確

• 心肌梗塞

– 症狀 – 心電圖

– CK-MB (肌酸酐激鋂-MB異構) – Troponin (心臟肌鈣蛋白)

– 核子醫學心臟掃描 – 心導管

(52)

Syndrome Diagnosis

• A: typical cases, B, C: atypical cases

(53)

Diagnosis in Practice

• Chief complaint

• Probability of symptoms and signs

• Probability of diagnostic tests

• Individualization

(54)

Diagnosis in Practice

• P(D+): 此疾病的盛行率

• P(T+): 此檢查陽性的機會

• P(T+|D+): 此檢查在此疾病中陽性的機會

(55)

Uncontrolled Experience

(56)

Evidence Level

• Deduction from theory

– 根據疾病機轉的理論，醫師認為有效的治療

• Uncontrolled Experience

– 根據經驗，醫師認為有效的治療

• Controlled Experience

– Well-controlled clinical research – Randomized clinical trial (RCT) – Evidence

(57)

Uncontrolled Experience in a Pre- scientific Era

• Scurvy and lemon juice

– 1747

• Cowpox and smallpox

– 1798

• Many alternative method

• Speculative theory

– Cause of disease

(58)

The Numerical Method

• Statistical method

– 1830

• Blood-letting and infectious disease

– Louis

(59)

The Era of Laboratory Research

• Myxoedema

– The same clinical picture after thyroidectomy – Tx with injection of thyroid extract

• Diabetes

– Tx with injection of insulin

• Pernicious anemia

– Tx with raw liver

– Tx with vitamin B12

(60)

The Spontaneous Course of the Disease

• Esophageal varices bleeding

– A research announced a new method can stop bleeding in up to 80-95%

– Spontaneous bleeding arrest?

(61)

Regression Towards the Mean

• 任何檢查都有變動性，造成誤差

– 正常受檢者 – 一次異常 – 複檢正常

(62)

Run of Luck

(63)

Bias

• Random errors

– Regression towards the mean – Runs of good or bad luck

• Bias (systemic errors)

– Uncontrolled clinical experience

(64)

The Placebo Effect

• Doctor-patient relationship

• Bio-psycho-social model

(65)

Alternative Medicine:

Pseudoscientific Thinking

• For all diseases?

– Chiropractic – Homeopathy

(66)

The Randomized Clinical Trial

(67)

Selection of Patients

• Entry criteria

– 診斷、疾病嚴重度、檢查結果…

• Exclusion criteria

– 懷孕、併有其他疾病、年齡…

• Informed consent (同意書)

• Story

– Rofecoxib

• Clinical Trial with lower risk patients

• Real patients have higher CV risk

Sampling Bias

(68)

Randomization

• New treatment for cancer

– 決定受試者使用哪種治療方法

• Double blinded trial

• Concealed allocation (隱瞞)

• Block randomization

– 多中心大型研究用之 – 幾人為一組

• Stratification

– 事先依據特定條件分層，如疾病嚴重度、病史

(69)

Choice of Treatment in the Control Group

• New treatment vs current treatment

• New treatment vs no treatment

– Use placebo

– Difficult to design in some studies

(70)

Principles of Blinding

• Unblinded trials bias

– 醫師事先知道分組 – 病人事先知道分組

• Double blinded

– 效果 – 副作用

• Single blinded

– 例: 手術

(71)

Cross-Over Trials

• 一位病人先後接受實驗組與對照組試驗

• 雙盲

• 除去不同病人的變異性

• 可減少收案數

• 避免carry-over effect

– 症狀改善與治療無關

(72)

Measures of Benefits and Harms

• Benefits

– 較容易偵測

• Harms

– Need large case number – Important

• Watch out all unexpected events for a new drug

• Report these to the relevant authority

(73)

Stopping Rules

• 收案數足夠

• Ethical reasons

– Treatment causes serious harms – 實驗組與對照組差異很大

– IRB (Institutional Review Board)

• 人體試驗委員會

• 臨床研究倫理委員會

(74)

Assessment of The Results

• Subjective judgment

• Bias

• Database

– Clean data

– Per protocol analysis

• 只包含完成試驗的個案

• 去掉drop-out的個案

• 有Overestimate問題

(75)

Assessment of The Results

• Intention-to-treat analysis

– 包含各種狀況個案

• 完成的

• Drop-out的

– 符合臨床實際狀況 – 例:

• Aspirin試驗 for Af and CVA

• H2-blocker and PPI for peptic ulcer

(76)

Statistical Analysis

• 比較兩種治療A與B

• Null hypothesis (H₀)

– A = B

• Alternative hypothesis (H_A)

– A <> B

• Type I error

– H

₀

為真，但實驗結果卻為A<>B

• Type II error

– H

_A

為真，但實驗結果卻為A=B

(77)

Statistical Analysis

• Significance test

– P-value

• Statistically significant

– P < 0.05

• 影響因素

– 分組比較的差異度 – 個案數

(78)

Systematic Reviews

• 優點

– 專家整理

– 許多研究結果之摘要

• 缺點

– 可能摻入作者個人觀點

• Publication bias

– 有統計意義的比較會被接受

• Cochrane Library

– Since 1993

– EBM (Evidence-based medicine)

(79)

From Trials to Practice

• 新藥上市

– RCT

• 不容易做RCT的狀況

– 未藥物治療組

• 用安慰劑

– 手術

– 物理治療

(80)

Factors in Decision

• Clinical data

• Patient personal condition

– Age

– Diseases

• Knowledge

– Evidence-based

• Economic

(81)

Medicine and the Humanities

(82)

The Four Components of Clinical

Reasoning

(83)

The Empathic-Hermeneutic Component

• Hermeneutic (解釋學) = art of interpretation

• “On Death and Dying” – Kubler-Ross

– Stages

• Denial

• Anger

• Bargaining

• Depression

• Acceptance

(84)

The Ethical Component

• Three kinds of norms (規範)

– Technical norms

• Scientific norms

– E.g. Good drugs

– Legal norms

• Abortion

– Ethical or moral norms

• A transplantation surgeon has only one organ but has two patients.

(85)

The Foundation of Health Care Ethics

• Three principles

– The Samaritan principle

• 急難救助免責

– The principle of distributive justice

• 公正分配資源

– The principle of autonomy

• 病人自主性

(86)

Clinical Research Ethics

• Helsinki Declaration of the World Medical Association

– 世界醫學會制定赫爾辛基宣言，做為進行人體研究時之倫理指導原則

– 人體試驗委員會(Institutional Review Board) – 知情同意(Informed Consent)

(87)

Clinical Research Ethics

• Research project

– Methodologically acceptable

• 科學性審查  對受試者及未來有益

– 利益衝突

• 廠商

• 研究設計

(88)

Critical Reading of Medical

Journals

(89)

Logical Analysis of Medical Papers

• 大部分研究在於建立因果關係

• Three approaches

– Forward

• 先假設原因，再收集其結果

• Conditional probability: P(effect | cause)

• Cohort studies

– Backward

• 找導致某結果的可能原因

• P(cause | effect)

• Some case-control studies

– Correlation (關聯性)

• 某時間點各種因子的關聯性

• Cross-sectional studies

• Risk factors or determinants are correlational and not

necessarily causal, because correlation does not prove causation

https://en.wikipedia.org/wiki/Risk_factor

(90)

Logical Analysis of Medical

Papers

(91)

Logical Analysis of Medical Papers

• The results of cohort studies are often

analyzed by determining the relative risk (RR，相對風險)

– 兩組的 P(effect | cause)比較

– 指暴露組的發病率或死亡率與非暴露組的發病率或死亡率的比值，適合作致病因子探討的指標

(92)

Logical Analysis of Medical Papers

• Odds ratio (OR) in case-controlled studies

– Odds

• 一件事情發生的機率與一件事情沒發生機率的比值

– OR

• 兩件事情的 odds 比較

(93)

Descriptive Statistics

• Mean

• SD

• Testing hypotheses

– P-value

(94)

Statistical Tests

• Fisher’s exact test (or chi-square test)

– 比較兩個比率

– 例: 兩個族群發生眼科病變的比率

• Two-sample rank sum test (Mann-Whitney U test)

– 比較ordinal scale的分布

• Two-sample t-test

– 比較interval scale的分布

(95)

Correlation Analysis

• Regression line

• Correlation coefficient: r

– 相關的強度

• P-value

(96)

Correlation Analysis

(97)

Life Table Analysis

(98)

Life Table Analysis

(99)

Life Table Analysis

• Used in

– Survival studies

– Recurrence of a disease

– Development of complication

– Metastasis of a malignant disease

(100)

Clinical Guideline

http://www.medicine.ox.ac.uk/bandolier/painres/download/whatis/

whatareclinguide.pdf

(101)

Protocol

• Protocols are rigid statements allowing little or no flexibility or variation.

• A protocol sets out a precise sequence of activities to be adhered to in the

management of a specific clinical condition.

• There is a logical sequence and precision of listed activities

(102)

Care Pathway

• Care pathways determine locally agreed, multidisciplinary practice, based on

guidelines and evidence where available, for a specific patient/client group.

• Care pathways form all or part of the

clinical record, document the care given and help to evaluate outcomes for continuous

quality monitoring

(103)

Guideline

• Guidelines reduce unacceptable or undesirable variations in practice and provide a focus for

discussion among health professionals and patients.

• They enable professionals from different disciplines to come to an agreement about treatment and devise a quality framework, against which practice can be measured.

• Guidelines can help commissioners and purchasers to make informed decisions and provide managers with a useful framework for assessing treatment costs.

(104)

Clinical guideline

• Clinical guidelines are systematically developed statements designed to help practitioners and patients decide on

appropriate healthcare for specific clinical conditions and/or circumstances

• Good guidelines can change clinical practice and influence patient outcome

(105)

Clinical guideline (2)

• The way in which guidelines are developed, implemented and monitored, influences the likelihood that they will be followed

• Guidelines should provide extensive,

critical and wellbalanced information on the benefits and limitations of various

diagnostic and therapeutic interventions so that the physician can carefully judge

individual cases

(106)

Clinical guideline (3)

• Applying guidelines to individual care is always likely to require judgment, even

when recommendations are properly linked to evidence

(107)

Good clinical guidelines

• Valid – leading to the results expected of them.

• Reproducible – if using the same evidence,

other guideline groups would come to the same results.

• Cost-effective – reducing the inappropriate use of resources.

• Representative/multidisciplinary – by involving key groups and their interests.

• Clinically applicable – patient populations affected should be unambiguously defined.

(108)

Good clinical guidelines

• Flexible – by identifying the expectations relating to recommendations as well as patient preferences.

• Clear – unambiguous language, which is readily understood by clinicians and patients, should be used.

• Reviewable – the date and process of review should be stated.

• Amenable to clinical audit – the guidelines should be capable of translation into explicit audit criteria.

(109)

Guideline Resources

• National Guideline Clearinghouse

– http://www.guideline.gov/

(110)

Example

• 2013 ACC/AHA Guideline on the

Treatment of Blood Cholesterol to Reduce Atherosclerotic Cardiovascular Risk in

Adults

– A Report of the American College of

Cardiology/American Heart Association Task Force on Practice Guidelines

(111)

Example

• 1.Introduction

• 2.Overview of the Guideline

– 2.1.Lifestyle as the Foundation for ASCVD Risk-Reduction Efforts

– 2.2.Initiation of Statin Therapy

• 3.Critical Questions and Conclusions

(112)

Example

• 4.Statin Treatment: Recommendations

– 4.1.Intensity of Statin Therapy in Primary and Secondary Prevention

– 4.2.LDL-C and Non–HDL-C Treatment Goals – 4.3.Secondary Prevention

– 4.4.Primary Prevention in Individuals 21 Years of Age With LDL-C 190 mg/dL

– 4.5.Primary Prevention in Individuals With Diabetes

– 4.6.Primary Prevention in Individuals Without Diabetes and With LDL-C 70 to 189 mg/dL

– 4.7.Risk Assessment in Primary Prevention – 4.8.Heart Failure and Hemodialysis

(113)

Example

• 5. Safety: Recommendations

• 6.Managing Statin Therapy: Recommendations

• 7.Selected Clinical and Population Subgroups

• 8.Limitations

• 9.Evidence Gaps and Future Research Needs

• 10.Conclusions

(114)