epi and fammed
TRANSCRIPT
1
วิ�ทยาการระบาดในงานเวิชศาสตร�ครอบคร�วิ
นพ. ส�ธี�ร� ร�ตนะมงคลก�ลภาควิ�ชาเวิชศาสตร�ป้�องก�นและส�งคม
มหาวิ�ทยาล�ยศร�นคร�นทรวิ�โรฒ
2
ห�วิข้%อบรรยาย• ควิามหมายข้องระบาดวิ�ทยา• การกระจายข้องโรค• ค'าจ'าก�ดควิามข้องโรค• Causal models• Screening• Prognostic factor• Epidemiological methods• Systematic reviews• Standardization• Measures in epidemiology
• Chance/ Random error• Bias• Confounder
3
Definition of EpidemiologyDefinition of Epidemiology
“The study of the distribution and determinants of disease frequency in human populations and the application of this study to control health problems”
Key Words in DefinitionKey Words in Definition • Disease frequency ‑ count cases, rate, records • Disease distribution ‑ who, when, where
– Frequency, distribution, other factors generate hypotheses about determinants
• A determinant is a characteristic that influences whether or not disease
occurs
4
การกระจายของโรค( Distribution )
• เวลาเวลา ( TIME ) ( TIME ) เมื่��อไร นานแค�ไหน เมื่��อไร นานแค�ไหน
• สถานที่�� สถานที่�� (PLACE) (PLACE) ที่��ไหนที่��ไหน
• บุ�คคล บุ�คคล (PERSON(PERSON )) อาย� เพศ อาชี�พ เชี�!อชีาติ# อาย� เพศ อาชี�พ เชี�!อชีาติ#
5
สถานการณ์%ไข&เล�อดออก(DHF+DF+DSS) เขติ 6 เปร�ยบุเที่�ยบุค�ามื่)ธยฐาน - 253943 2544( ), , มื่ค - . พย.2545(ส)ปดาหที่��48)
229 114 15
69616
6114
70
71
1026
1327
613
1251
2243
1240
598428
196
4641055
10811363
95241
26 23
613
285326368
166 195543
987
2061
3062
02505007501000125015001750200022502500275030003250
มื่ค กพ มื่�ค เมื่ย พค มื่#ย กค สค กย ติค พย ธค
MED(39-43)
2544
2545
จ-านวน(ราย)
11719/26
Time
6
Stomach cancer mortality rates: Migration study
58.4
29.9
11.7
8
0 10 20 30 40 50 60 70
Japanese in Japan
JapaneseImmigrants to
California
Sons of JapaneseImmigrants
Native Californians(Caucasians)
Mortality rates (per 100,000)
7
Population Pyramid
8
ระบาดวิ�ทยาสถานท�)ท�)เส�ยช�วิ�ตข้องผู้+%ป้,วิยมะเร-งท�)
ร�ฐโนวิาสโกเท�ย แคนนาดา
Conclusion: “Over time, more patients with cancer, especially women, elderly people and people with longer survival after diagnosis, died outside of hospital in Nova Scotia.”
9
10
11
Proportion of deaths out of hospital and in hospital among adults in Nova Scotia who died of cancer, 1992–1997.
CMAJ. 2003 February 4; 168(3): 265–270.
12
Home visit ก�บการลดการเส�ยช�วิ�ตในโรงพยาบาล
13
14
Natural Progression in Natural Progression in Epidemiologic ReasoningEpidemiologic Reasoning
1st – Suspicion that a factor influences disease occurrence. Arises from clinical practice, lab research, examining disease patterns by person, place and time, prior epidemiologic studies
2nd – Formulation of a specific hypothesis
3rd – Conduct epidemiologic study to determine the relationship between the exposure and the disease. Need to consider chance, bias, confounding when interpreting the study results.
4th – Judge whether association may be causal. Need to consider other research, strength of association, time directionality
15
ค'าจ'าก�ดควิามข้องโรค(Case Definition)
16
Case definition
Criteria
• Epidemiologic criteria
• Clinical criteria
• Diagnostic criteria
• Exclusion criteria
17
18
19
Measuring Occurrence of Diseases
20
• อ)ติรา (Rate) = ติ)วติ)!ง ติ)วหาร
– ความื่ชี�ก (Prevalence Rate) อ)ติราป.วยรายเก�า+
รายใหมื่�
– อ�บุ)ติ#การ (Incidence Rate) อ)ติราป.วยรายรายใหมื่�เที่�าน)!น
– อ)ติราติาย (Mortality Rate) อ)ติราติายในประชีากร
– อ)ติราติายจากโรค (Fatality Rate ) อ)ติราติายในคนที่��ป.วย
การว)ดการเก#ดของโรค
21
Incidence densityCumulative incidence
Suthee Rattanamongkolgul
22
JAN1995
JAN1996
APR1995
JUL1995
OCT1995
DeathCuredDisease
1234
5
6
78910
Normal
1212
129
3
6
6
23
มื่� 2 แนวที่างในการว)ด
1) อ�บุ)ติ#การสะสมื่ Cumulative incidence
= จ-านวนผู้1&ป.วยรายใหมื่�ในชี�วงเวลาที่��ก-าหนด
ประชีากรที่��เส��ยงในชี�วงเวลาน)!น= 40 = 1.25 /1,000
32,000
X 10(n)
การว)ด อ�บุ)ติ#การของการเก#ดโรค
24
2) อ)ติราอ�บุ)ติ#การ (Incidence density or Incidence rate )
• “ ” การเพ#�มื่ มื่#ติ#ของเวลา ลงไปในติ)ว “หาร Person-time”
• Person-month, Person-year• 1 Person-year = Following 1 person for 1 year period
• 10 Person-year = Following 1 person for 10 year period
= Following 10 persons for 1 year period
การว)ด อ�บุ)ติ#การของการเก#ดโรค
25
•2) อ)ติราอ�บุ)ติ#การ Incidence density or Incidence rate = จ-านวนผู้1&ป.วยรายใหมื่�ในชี�วง
เวลาที่��ก-าหนด จ-านวน Person-years of ของการติ#ดติามื่ในชี�วงไมื่�เก#ดโรค
• หากติ#ดติามื่คน 100 คนในเวลา 1 ป2และพบุ ว�า 20 คนเก#ดโรค อ)ติราอ�บุ)ติ#การค�อ
• 20 cases/100 person-years การติ#ดติามื่
• 20 = 20 / 100 person-years
100 person-years
X 10X 10(n)(n)
การว)ด อ�บุ)ติ#การของการเก#ดโรค
26
Smoking No. of stroke Person-
years Incidence rate
of observation
/100,000 person-years
Never 70 395,594
17.7
Ex-smoker 65 232,712
27.9
Smoker 139 280,141
49.6
Total 274 908,477
30.2
ความื่ส)มื่พ)นธ%ระหว�างการส1บุบุ�หร��ก)บุอ�บุ)ติ#การ
การเก#ดอ)มื่พาติ ของประชีากร 118,539 คนในเวลา 8 ป2
27
Epidemiological methods
28
InducInductive tive apprapproachoach
HypothesisHypothesis
ObservationObservation
DeduDeductive ctive approapproachach
การสร&างสมื่มื่�ติ#ฐาน
การที่ดสอบุสมื่มื่�ติ#ฐาน
การก#นหมื่ากอาจเป3นสาเหติ�ของมื่ะเร4งชี�องปาก
มื่ะเร4งที่��ชี�องปากส�วนใหญ่�มื่)กก#นหมื่าก
การศ6กษาเชี#งพรรณ์นา (Descriptive epidemiology)
การศ6กษาเชี#งว#เคราะห% (Analytic epidemiology)
29
Observational
Intervention
Descriptive
Analytical
Clinical Trials
Case reportsCase series
Ecological Studies
Cross Sectional Survey
Cohort Study
Case-Control Study
Study Designs Clinical Research
Population
Individuals
Field Trials
Community Trials•Equivalent trial•Non-inferior trial•Superior trial
•Time trend•Migration study•Geographical correlation
•Case-control•Nested-case control
•Cohort•Historical cohort
•Prevalence study•Analytical cross-sectional
Generating hypothesis
Testing hypothesis
Experimental study
Observational study
Longitudinal
30
Intervention studyExperimental study
Clinical trial
31
RAMDOMIZATION
Eligibility criteria
TARGET POPULATION
EXPERIMENTAL POPULATION
Unsuitable subjects
Informed consentRefusals
Intervention group
STUDY POPULATION
Control group
Follow-up
OutcomeOutcome
Follow-up
Experimental Studies
32
Types of Intervention studies
Types Purposes Subjects
Clinical Trials Treatment Individuals
Field Trials Prevention Individuals
Community Trials Prevention Community
33
Concepts• Phase I, II, III, and IV• Comparison:
– historical control, before and after, parallel • Improvements unrelated to treatment
– Contamination, co-intervention, regression to the means• Randomization
– Sequence generation + sequence concealment• Admissibility criteria
– Uniform characteristics, comparability via homogeneity• Outcome ascertainment
– Blinded • Ethics
– Informed consent, ethic guideline
34
Strengths• Randomisation ensures that
on average known and unknown confounders are equally distributed between intervention and control groups
• Experimental, hence control over intervention, circumstances and subjects
• Can study multiple effects from single exposures
• Clear temporal relationship can be established
Weaknesses
• May be expensive, time-consuming and logistically difficult (especially if prolonged follow-up)
• Generalisability from trial setting to everyday practice may be problematic
• Less useful for rare outcomes
• Loss to follow may affect validity of results
• Lack of blinding may introduce bias
• Ethical problems
35
Cohort studyHistorical/Retrospective cohort study
36
Cohort study
SampleNon-participants
Non-exposedExposed
Lost to follow-up
Measure outcome Measure outcome
Self Selection
No outcome at the beginning of study
37
Smokers Non-smokers
Incidence rate of
lung cancer
Incidence rate of
lung cancer
Cohort study
Folllow up Folllow up
38
a
c+d
a+b
a+c
c
Nb+d
Y N
Y
N
Lung cancer
Smoking
Incidence exposed = a/(a+b)Incidence not exposed = c/(c+d)Relative Risk = [a / (a+b)] / [c / (c+d)]
b
d
39
40
Cohort study and Relative Risk
100
50,000
10,000
150
50
N58,950
Y N
Y
N
Lung cancer
Smoking
Incidence exposed (Ie) = a/a+b = 100/10,000 = 0.01 = 10/1,000Incidence not exposed (Io)= c/c+d = 50/50,000 = 0.001= 1/1,000 Relative Risk = Ie/Io = 0.01 / 0.001 = 10
9,900
49,050
a
c+d
a+b
a+c
c
Nb+d
Y N
Y
N
b
d
41
Strengths and weaknesses of prospective cohort studies
Strengths
• Prospective studies minimise bias in exposure measurement and may allow multiple measurements, increasing precision
• Can study multiple effects from single exposures
• Clear temporal relationship can be established
• Good for rare exposures
Weaknesses
• Expensive, time-consuming and logistically difficult (especially prospective studies in diseases with long latency)
• Less useful for rare diseases
• Loss to follow may affect validity of results
42
Historical CohortRetrospective Cohort
• Cohort data can be collected prospectively (as events occur over time) or retrospectively (historical records)
• Dyestuff workers and bladder cancer– Employment records and vital statistic used to
capture events from the past– On-the-job exposure to aniline dyes was
associated with a large increase in bladder cancer
43
Case-control studyNested case-control study
44
People with liver cancer
[Case]
People with no liver cancer[Control]
History of hepatitis B carrier
History of hepatitis B carrier
Case-control study
Ask Ask
45
Case Control Studies
Strengths
• Relatively quick and inexpensive
• Can be used for rare diseases
• Can study multiple aetiological factors of single diseases
Weaknesses
• Less useful for rare exposures
• Particularly prone to bias (especially selection and recall bias)
• Cannot compute incidence rates (unless population based)
46
Nested Case-Control StudiesFigure 3
Study PopulationTIME 1
YEARS
TIME 2DevelopDisease
Do NotDevelopDisease
CASES CONTROLS
CASE-CONTROL STUDY
Obtain interviews, bloods, urines, etc.
47
A. Advantages of Nested Case-Control Studies
1. Possibility of recall bias is eliminated, since data on exposure are obtained before disease develops.
2. Exposure data are more likely to represent the pre-illness state since they are obtained years before clinical illness is diagnosed.
3. Costs are reduced compared to those of a prospective study, since laboratory tests need to be done only on specimens from subjects who are later chosen as cases or as controls.
48
Predictors of Follow-up of Atypical and ASCUS Papanicolaou Test Results in a High-Risk Population
M. Diane McKee, MD, MS; Clyde Schechter, MA, MD; William Burton, PhD; Michael Mulvihill, DrPH
July 2001 · Vol. 50, No. 7 JFP
Online
49
Prevalence, incidence, morbidity and treatment patterns in a cohort of patients diagnosed with
anxiety in UK primary care.• Fam Pract. 2010 Feb;27(1):9-16. Epub 2009 Nov 1. Martín-Merino E, Ruigómez A,
Wallander MA, Johansson S, García-Rodríguez LA.
METHODS: • The Health Improvement Network was used to identify all patients aged 10-79 years
with • a new diagnosis of anxiety in 2002-04 (n = 40 873) and • age-, sex- and calendar-year-matched controls (n = 50 000). • A nested case-control analysis was used to quantify potential risk factors for anxiety
by multivariate logistic regression.RESULTS: • The prevalence of anxiety was 7.2% and • the incidence was 9.7 per 1000 person-years. • Incidence and prevalence were highest in women and young adults (20-29 years). • Anxiety was associated with heavy alcohol use, smoking and addiction problems as
well as stress, sleep and depression disorders. • Anxiety patients used health care services more frequently than controls. • Among patients diagnosed with anxiety, 63% were treated pharmacologically.
Antidepressants accounted for almost 80% of prescriptions.
50
Case-control study study
Liver cancer
No liver cancer
HBsAg+ 400a.
100b.
HBsAg- 200c.
600d.
Odd ratio = = = 400/100
200/600
400 x 600
100 x 200
Odd in smokers
Odd in non-smokers
Odd ratio (OR) = 12 =a/b
c/d
51
Odd ratio (OR) and relative risk (RR)
• Relative risk and odd ratio have the same interpretation
• OR and RR are ususally presented with 95% confience intervals (CI) to indicate possible value of OR or RR within possibility of 95%
e.g. OR = 2.1(95%CI:1.2-3.6)
OR = 1.5(95%CI: 0.9-2.3)
OR = 0.5 (95%CI:0.2-0.8)
52
X X ค0อ ค1าเฉล�)ยข้องต�วิอย1าง ค0อ ค1าเฉล�)ยข้องต�วิอย1าง ควิามแป้รป้รวิน ควิามแป้รป้รวิน , Variance , Variance (S(S2))
A
Probability
53
Ecological study
54
Types of ecological study
• Time trends
• Ecological Correlation or Geographical studies)
• Migrant studies
55
Time trend
Incidence rate of breast cancer in US
0
20
40
60
80
100
120
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
Year
Inci
den
ce r
ate
Time trends• Trends over time can give clues to aetiology• Can sometimes be linked to trends in exposure• Need to consider changing definitions over time
56
57
58
59
60
Ecological study
Female Literacy %
100806040200
120
100
80
60
40
20
0
US
ROM
PR
PER
PAN
NIC
NEP
MON
MAL
KEN
J
A
IND
GUA
GAM
K
CHI
USA
BOL
BAN
T
Infant Mortality
Infant mortality rates
61
Strengths and Weaknesses
Strengths
• Simple and cheap
• Often use routinely available information
• Good first step - hypothesis generating
• Assess effect of exposures on outcomes at the population rather than individual level
Weaknesses
• Ecological fallacy - averaging of exposure and outcome information at population level -> cannot link exposure and disease in individuals
• Lack of information on confounders
• Limitations of routine data sources – possible inaccuracy and biased comparisons
• May miss sub-groups effects
62
Cross-sectional study
63
Cross-sectional studies
• also called survey or prevalence study • Survey of group to identify prevalence of disease or
exposure• Exposure and disease status assessed at the same time• Individual is unit of observation and analysis• Typically descriptive in nature to quantify magnitude of
the problem• used to generate hypotheses
64
Does smoking cause depression? Or does depression cause smoking?
School Sample
Identify SmokersIdentify mood
disturbed students
Findings: strong association between mooddisturbance, especially depression and smoking
Crossectional study
65
1120180
68020
เป3น ไมื่�เป3น
โรคห)วใจ
ส1บุ
ไมื่�ส1บุการส1บุบุ�หร��700
-
2000
รวมื่
1300
200 1800
ส)ดส�วนของการส1บุบุ�หร�� ในคนที่��คน เป3นโรคห)วใจ =
ส)ดส�วนของการส1บุบุ�หร��ในคนที่�ไมื่�เป3น โรคห)วใจ =
รวมื่
680/1800 = 38%
20/200 = 10%
66
Strengths• Magnitude of
problems
• May generate new etiologic hypotheses
Weaknesses• No cause-effect
• No temporality
• Prevalent cases are survivors
Cross sectional studies
67
Case-seriesCase report
68
69
70
71
INVASIVE DEVICES USE in Patients with MRSA (16 cases)
Procedure Frequency
IV cath.NG tubeET tubeVentilatorTracheostomyCut downFoley’s Cath.Others
1611887761
Sensitivity Frequency
VancomycinChloramphenicolFosfomycin
161410
Drug sensitivity (16 cases)
MRSA infection in HRH Maha Chakri Sirindhorn Medical Center during the last 2 years By Brian Lee, Chayanant Siwarungson
Case-series studyCase-series study
72
Case report/study Case report: CHD as a potential cause of chest pain in adult patients.
Tsuang W, Ahmed K, Eckert D, Helmy T, Grunenwald P.
Am Fam Physician. 2008 Oct 15;78(8):911, 914. :
Case report: amelanotic melanoma located on the lower extremity.
Valdes A, Kulekowskis AM, Curtis L.
Am Fam Physician. 2007 Dec 1;76(11):1614, 1618.
Case report: Patient adherence to drug regimens vital to treatment.
Simmons BB, Dubreuil AL.
Am Fam Physician. 2007 Sep 15;76(6):769-70. :
Case report: differentiating artifact from true ventricular tachycardia.
Mascitelli L, Pezzetta F.
Am Fam Physician. 2006 Sep 15;74(6):921. :
Sleep disturbances in the disabled child--a case report and literature review.
Adlington K, Liu AJ, Nanan R.
Aust Fam Physician. 2006 Sep;35(9):711-5.
73
Herpes encephalitis presenting as mild aphasia: case report.Khan OA, Ramsay A.BMC Fam Pract. 2006 Mar 24;7:22.Case report: expanding the differential diagnosis of
intractable cough.Shah RK, Ditkoff M, Karmody CS.Am Fam Physician. 2005 Nov 15;72(10):1975-6. : Case report: use of prazosin for treatment of
posttraumatic stress disorder.Griffith LJ.Am Fam Physician. 2005 Sep 1;72(5):758, 761. : The role for vitamin B-6 as treatment for
depression: a systematic review.Williams AL, Cotter A, Sabina A, Girard C, Goodman
J, Katz DL.Fam Pract. 2005 Oct;22(5):532-7. Epub 2005 Jun
17. : Case report: Urticaria following intentional
ingestion of cicadas.Piatt JD.Am Fam Physician. 2005 Jun 1;71(11):2048, 2050. Case report: insect bite reveals botfly myiasis in an
older woman.Sajjad N, Biederman G.Am Fam Physician. 2005 Apr 1;71(7):1262-3. : Case report: Late-onset eclampsia presents as
bilateral cortical blindness.Gold KJ, Barnes C, Lalley J, Schwenk TL.Am Fam Physician. 2005 Mar 1;71(5):856, 858, 861. A case report: ethics of a proposed qualitative
study of hospital closure in an Australian rural community.
Fraser J.Fam Pract. 2004 Feb;21(1):87-91.Hypermagnesemia. Elderly over-the-counter drug
users at risk.Fung MC, Weintraub M, Bowen DL.Arch Fam Med. 1995 Aug;4(8):718-23. : 7620603
[PubMed - Detailed description of a successful outpatient
taper of phenobarbital therapy.Geurian K, Burns I.Arch Fam Med. 1994 May;3(5):458-60.PMID: 8032508
[PubMed - indexed for Hyponatremia due to enalapril in an elderly patient.
A case report.Gonzalez-Martinez H, Gaspard JJ, Espino DV.Arch Fam Med. 1993 Jul;2(7):791-3.PMID: 8111505
[PubMed - indexed for
74
75
Case Report: Subarachnoid Hemorrhage in Woman with Migraines
• Am Fam Physician. 2009 Mar 1;79(5):374-376.� � �• TO THE EDITOR: • A 45-year-old woman with a history of migraines and a recent
diagnosis of hypertension presented with a two-day history of increasing headache.
• She had taken over-the-counter medications with some relief. On the second day, the patient awoke with terrible pain located in the occiput and radiating to the forehead, associated with photophobia, nausea, and vomiting.
• The headache was unlike any she had ever experienced. Physical examination revealed a dehydrated, distressed woman with a blood pressure of 180/120 mm Hg, pulse of 110 beats per minute, and body temperature of 97 F (36.1 C). � �
• There were no neurologic deficits. A computed tomography (CT) scan of the head showed a subarachnoid hemorrhage
76
• Computed tomography scan of the brain post aneurysm repair surgery, status post left frontoparietal craniotomy, showing boneflap (yellow arrow) and suprasellar aneurysm clips (green arrow). There is high-attenuation blood in the interhemispheric fissure showing subarachnoid hemorrhage (white arrow), and progressive generalized cerebral edema with diffuse attenuation of cerebral sulci (red arrow). Acute ischemia involving the medial frontal and parietal lobes (purple arrow) is likely the result of compromise of the anterior cerebral artery bilaterally, perhaps related to vasospasm.
77
Case report and case series
ข&อด�• ค&นพบุโรคใหมื่� ๆ• เป3นพ�!นฐานในการสร&างสมื่มื่�ติ#ฐานเก��ยวก)บุป9จจ)ย
เส��ยง
ข&อด&อย• Case report มื่าจาก ผู้1&ป.วยรายเด�ยวการน-า
ผู้ลไปใชี&ก)บุผู้1&ป.วยรายอ��นอาจใชี&ไมื่�ได& เชี�น ผู้ลของการร)กษาด&วยสมื่�นไพร
• Case series แมื่&มื่าจากผู้1&ป.วยหลายรายแติ�แติ�ย)งไมื่�ได&พ#ส1จน%สมื่มื่�ติ#ฐานของสาเหติ�
78
Systematic reviewsMeta-analysis
79
Systematic review and meta-analysis
• Systematic review – aims to identify, evaluate and interpret all the relevant available research evidence for a particular question.
• Meta-analysis – process of using statistical methods to combine quantitatively the results of different studies
80
81
82
83
84
85
Elements of a study protocol for a systematic review and meta-analysis
ObjectivesBackgroundInformation retrievalData collectionData analysis1. Create summary data2. Examine heterogeneity3. Examine publication bias4. Consider conducting subgroup analysis
(according to a priori hypotheses)
86
Funnel plot to detect publication bias
Egger et al. Systematic reviews in health care. London: BMJ books, 2001.
87
Hierarchy of study design
• Systematic reviews and meta-analysis• Randomised controlled trial• Non-randomised trial
– Controlled trial– natural experiment– before and after study etc
• Cohort study• Case control study• Cross sectional study• Ecological study• Case series• Case study/report
Experimental
Observational
Analytical
Descriptive
88
In practice, choice of study design depends on:
• State of knowledge
• Frequency of exposure and disease
• Time, cost and other feasibility considerations
• Each study design has unique and complementary advantages and disadvantages
89
General model of causation
Sufficient causeComponent causeNecessary cause
90
GENERAL MODEL OF CAUSATION (CAUSAL PIES)
This illustration shows a disease that has 3 sufficient causal complexes, each having 5 component causes. A is a necessary cause since it appears as a member of each sufficient cause. B, C, and F are not necessary causes since they fail to appear in all 3 sufficient causes.
91
GENERAL MODEL OF CAUSATION (CAUSAL PIES) BY KJ ROTHMAN
Sufficient cause: • A set of conditions without any one of
which the disease would not have occurred. (This is one whole pie.)
92
GENERAL MODEL OF CAUSATION (CAUSAL PIES) BY KJ ROTHMAN
Component cause:
• Any one of the set of conditions which are necessary for the completion of a sufficient cause. (This is a piece of the pie.)
93
GENERAL MODEL OF CAUSATION (CAUSAL PIES) BY KJ ROTHMAN
Necessary cause: • A component cause that is a member
of every sufficient cause.
94
Attributes of the causal pie
1. Completion of a sufficient cause is synonymous with occurrence (although not necessarily diagnosis) of disease.
2. Component causes can act far apart in time. 3. A component cause can involve the
presence of a causative exposure or the lack of a preventive exposure.
4. Blocking the action of any component cause prevents the completion of the sufficient cause and therefore prevents the disease by that pathway
95
Causal "guidelines" suggested by Sir AB Hill (1965)
Purpose: Guidelines to help determine if associations are causal.Should not be used as rigid criteria to be followed slavishly. Hill even stated that he did not intend for these "viewpoints" to be used as “hard and fast rules.”
96
Causal "guidelines" suggested by Sir AB Hill (1965)
1. Strength of the association2. Consistency3. Specificity4. Temporality5. Biological gradient6. Plausibility7. Coherence8. Experiment9. Analogy
97
เกณ์ฑ์%การติ)ดส#นการเป3นเหติ�และผู้ล• Magnitude of effect ข้นาดข้องควิามส�มพ�นธี�• Temporal relationship การเก�ดสาเหต�ก1อนการเก�ด
โรค• C onsistency ควิามสอดคล%องข้องการศ3กษาใน
ป้ระชากรหลาย ๆ กล�1ม• - Dose response relationship การเพ�)มข้37นหร0อลด
ลงข้องสาเหต�ม� ผู้ลกระทบตอ1การเก�ดโรค• Plausibility การอธี�บายได%ด%วิยหล�กวิ�ทยาศาสตร�• Reversibility การลดป้9จจ�ยสาเหต� แล%วิท'าให%การเก�ด
โรคลดลง
98
การวิ�ดผู้ลกระทบข้องป้9จจ�ยเส�)ยงข้องการเก�ดโรค
(Measures of Impact)
Attributable risk
Population attributable risk
Lyle Petersen 1999, Thomas Grein 2000-2004, Marta Valenciano 2005
99
• Cohort study to examine causes for automobile-related deaths
ถ&าที่�านเป3น รมื่ติ.กระที่รวงสาธารณ์ส�ขซึ่6�งติ&องการลดการติายจากอ�บุ)ติ#เหติ�ที่างถนนด&วยงบุประมื่าณ์ที่��จ-าก)ดที่��จะรณ์รงค%ในเร��องในเพ��อให&เก#ดผู้ลกระที่บุที่��ส1งส�ดระหว�างการเมื่าแล&วข)บุหร�อการข)บุรถเร4วในประชีากรแห�งหน6�ง
100
RR = 5.0
Incidence exposed
Incidence unexposed
Risk difference
Relative risk
0.0005 0.0001 0.0004 5.0
0.005 0.001 0.004 5.0
0.05 0.01 0.04 5.0
0.5 0.1 0.4 5.0
Relative risk ไม1สามารถแสดงผู้ลกระทบข้องexposure ท�)ม�ในป้ระชากรได%
101
Measures of Impact
• Measures of association providing information about absolute effects of exposure
• Reflect apparent contribution of an exposure to the frequency of disease
• Two concepts– Attributable risk among exposed– Population attributable risk
102
1. Impact measures among exposed
2. Impact measures in the population
103
Attributable Risk (AR)
• Quantifies disease burden in exposed group attributable to exposure
• Provides answers to– what is the risk attributed to the exposure?– what is the excess risk due to the exposure?
• Calculated as risk difference (RD)
104
Attributable Risk
Incidence
Exposed Unexposed
Iexposed – Iunexposed
I = Incidence
105
Dead Not dead Risk RD
Fast 100 1900 2000 0.05
0.04
Slow 80 7920 8000 0.01
180 9820 10000
A cohort study of accidental death due to fast driving
106
Dead Not dead Risk RD
Fast 100 1900 2000 0.05
0.04
Slow 80 7920 8000 0.01
180 9820 10000
A cohort study of accidental death due to fast driving
107
Drunk 45 255 300 0.150
0.136
Not d. 135 9565 9700 0.014
180 9820 10000
A cohort study of accidental death due to drunk driving
Dead Not dead Risk RD
108
Drunk 45 255 300 0.150
0.136
Not d. 135 9565 9700 0.014
180 9820 10000
A cohort study of accidental death due to drunk driving
Dead Not dead Risk RD
109
Attributable Risk Percent
Incidence
Exposed Unexposed
%exposed
unexposedexposed
I
I - I100 x
RR
1 -RR x100
110
Dead Not dead Risk Attributable Risk %
Fast 100 1900 2000 0.0500.050- 0.010 0.050
= 80%Slow 80 7920 8000 0.010
180 9820 10000
A cohort study of accidental death due to fast driving
AR% = [Ie – Io] x100 Ie
RR% = Ie = 0.050
Io 0.010
111
Dead Not dead Risk Attributable Risk %
Drunk 45 255 300 0.150
Not d. 135 9565 9700 0.014
180 9820 10000
A cohort study of accidental death due to drunk driving
0.150- 0.014 0.150
= 91%
AR% = [Ie – Io] x100 Ie
RR% = Ie = 0.150
Io 0.014
112
Prevented Fraction (PF)• For exposures associated
with decreased riskAttributable risk (AR) is called Prevented fraction
(PF)
• If relative risk <1– proportion of potential cases which would have
occurred if the exposure had been absent– proportion of potential cases prevented by the
exposure– E.g. in the case of vaccination for disease
prevention
113
Prevented fraction
RR -1
I
I -I PF
unexposed
exposed unexposed
114
Prevented Fraction Percent (PF%)
Incidence
Exposed Unexposed
%unexposed
exposedunexposed
I
I - I x100RR)-(1x100
Unexposed Exposed
115
PF: Vaccine efficacy
Pop. Cases Cases/1000 RR
Vaccinated 301,545 150 0.49 0.28
Unvaccinated 298,655 515 1.72 Ref.
Total 600,200 665 1.11
0.72 0.28 - 1
0.72 1.72
0.49 - 1.72 PF
116
1. Impact measures among exposed
2. Impact measures in the population
117
Population Attributable Risk
Risk
Population Unexposed
unexposed population I -I
118
Population Attributable Risk Percent (PAR%)
100 x 1 1)-(RR P
1) -(RR P PAR%
100 x I
I - I PAR%
population
unexposedpopulation
where P = % population exposed
119
Dead Not dead Risk
Fast 100 1900 2000 0.050
Slow 80 7920 8000 0.010
180 9820 10000 0.018
PAR%: Fast driving
44% 100 x 0.018
0.010 - 0.018 PAR%
0.008 0.010 - 0.018 PAR
120
Dead Not dead Risk
Drunk 45 255 300 0.150
Not d. 135 9565 9700 0.014
180 9820 10000 0.018
PAR%: Drunk driving
22% 100 x 0.018
0.014 - 0.018 PAR%
0.004 0.014 - 0.018 PAR
121
Conclude
• Driving related deaths in population
– 44% presumably due to fast driving
– 22% presumably due to drunk driving
122
PAR% in USA and Italy for cervical cancer,
various risk factors
Source, Parazzini et al. 1990
123
Direct standardizationIndirect standardization
การเป้ร�ยบเท�ยบ incidence rate หร0อ mortality rate ระหวิ1าง
ป้ระชากร หร0อระหวิ1างกล�1มป้ระชากรท�)สนใจก�บป้ระชากรอ%างอ�ง
124
อ�ตราการตายท�)ส+งข้องร�ฐฟอร�ดา
125
126
127
128
กล�1มอาย�
ส)ดส�วนประชีากรจ)งหว)ด
ส)ดส�วนประชีากร
อ-าเภอ
อ)ติราการเป3น อ)มื่พาติติ�อ
1000
จ)งหว)ด
อ)ติราการเป3น อ)มื่พาติติ�อ
1000
อ-าเภอ15-34 30% 24% 3 3
35-44 28% 22% 8 6
45-59 22% 27% 14 12
60+ 20% 27% 23 21
รวมื่ 100% 100% 11 11
ส)ดส�วนประชีากรและอ)ติราการเป3นโรคห)วใจติ�อ 1000 ในระด)บุจ)งหว)ดและอ-าเภอ
129
กล�1มอาย� ส)ดส�วนประชีากรจ)งหว)ด
(A)
อ)ติราการเป3น อ)มื่พาติติ�อ 1000
ในอ-าเภอ(B)
A*B
15-34 0.30 3 0.9/1000
35-44 0.28 6 1.68/1000
45-59 0.22 12 2.64/1000
60+ 0.20 21 4.2/1000
รวมื่ 1.00 11 9.42/1000
DIRECT STANDARDIZATION
130
เม0)อเป้ร�ยบเท�ยบแล%วิพบวิ1าage-standardized MI mortality rate
ของอ-าเภอเที่�าก)บุ 9.42/1,000ซึ่6�งติ-�ากว�าอ)ติรา
ของจ)งหว)ดซึ่6�งมื่�ค�าเที่�าก)บุ 11.0/1,000
การที่-า age-standardized rate ใชี&ได&ก)บุที่)!งmortality และ incidence
มื่)กใชี&เปร�ยบุเที่�ยบุค�า incidence หร�อmortality ระหว�างประชีากร เชี�น incidence rate ของ โรคมื่ะเร4ง โดยเร�ยกว�า adjusted rate โดยใชี&ประชีากรโลกเป3นประชีากรอ&างอ#ง
131
กล�1มอาย� ประชีากรอ-าเภอ(C)
อ)ติราติายของ จ)งหว)ดติ�อ 1000
(D)
C*D
15-34 240 3 0.72
35-44 220 8 1.76
45-59 270 14 3.79
60+ 270 23 6.21
รวมื่ 1,000 11 12.47
Standardized mortality ratio (SMR)= actual deaths/expected
deaths= 11/12.47
= 0.88 หร�อ ค#ดเป3น 88%
INDIRECT STANDARDIZATION
132
Standardized mortality ratio (SMR)= actual deaths/expected deaths= 11/12.47= 0.88 หร�อ ค#ดเป3น 88%SMR = 1 , การติายในอ-าเภอเที่�าก)บุจ)งหว)ดSMR < 1 , การติายในอ-าเภอน&อยกว�าจ)งหว)ดSMR > 1 , การติายในอ-าเภอมื่ากกว�าจ)งหว)ด
การที่-า indirect standardisation ใชี&ได&ก)บุที่)!งincidence และ mortality
133
Number Needed to TreatNumber Needed to Harm
134
การแสดงผู้ลกระทบข้องการได%ร�บ exposure ต1อการเก�ดโรคเป้;นค1าควิามเส�)ยงม�ควิามยากต1อการ
ป้ระเม�นหร0อเห-นภาพต1อผู้ลกระทบ จ3งม�การแสดง ผู้ลกระทบข้องการได%ร�บ exposure ซึ่3)งอย+1ในร+ป้
ข้อง intervention เป้;น จ'านวินผู้+%ป้,วิยท�7งหมดท�)จะผู้1านการร�กษา ซึ่3)งจะได%ร�บ
ผู้ลข้องการป้�องก�นโรค 1 คน เร�ยกวิ1า number needed to treat (NNT)
จ'านวินผู้+%ป้,วิยท�7งหมดท�)จะผู้1านการร�กษา ซึ่3)งจะเก�ด ภาวิะแทรกซึ่%อนจากการร�กษา 1 คน เร�ยกวิ1า
number needed to harm (NNH)
135
โรคไข้%หวิ�ด ไม1เป้;นโรคไข้%หวิ�ด
Totals
ฉ�ดวิ�คซึ่�น(experim
ent)
1 29 30
ไม1ฉ�ดวิ�คซึ่�น(control)
9 21 30
Totals 10 50 60
30.0219
9
dc
cCER
033.0291
1
ba
aEER
ควิามเส�)ยงเป้;ฯโรคไข้%หวิ�ด
ในกล�1มได%วิ�คซึ่�น
ควิามเส�)ยงเป้;ฯโรคไข้%หวิ�ดในกล�1มไม1ได%วิ�คซึ่�น
136
• เพราะฉะน�7น Absolute risk reduction (ARR)
• = |EER-CER| = |0.033-0.30| = 0.267
• NNT =
745.3267.0
1
ARR
1
137
• การเป้ร�ยบเท�ยบแอสไพร�นก�บยาหลอกเพ0)อ ป้�องก�นการเก�ด myocardial infarction
• ม�ค1า ARR เท1าก�บ 0.04
• ด�งน�7นค1า NNT จะเท1าก�บ 25
• น�)นค0อ ถ%าให%แอสไพร�นแก1ผู้+%ป้,วิยจ'านวิน 25 ราย จะ สามารถป้�องก�นการเก�ด myocardial infarction
ได% 1 ราย
138
• ในท'านองเด�ยวิก�น การให%ยาก�บผู้+%ป้,วิยอาจท'าให%เก�ดผู้ลอ0)น ท�)ไม1ต%องการและอาจเป้;นอ�นตรายต1อผู้+%ป้,วิยได%
• ค1า NNT ท�)ค'านวิณได%อาจเร�ยกวิ1า number needed to harm (NNH)
• ต�วิอย1างเช1น การใช%แอสไพร�นท�)ท'าให%ผู้+%ป้,วิยม�ควิามเส�)ยง ต1อการม�เล0อดออก ถ%า ARI ม�ค1าเท1าก�บ 020. จะได%
NNH เท1าก�บ 5 (1/0.2 ) น�)นค0อ ในผู้+%ป้,วิยท�)ใช%ยา แอสไพร�นท�กๆ 5 รายจะม�ผู้+%ป้,วิย 1 รายท�)ม�เล0อดออก2.0
1
139
• ค0อ จ'านวินผู้+%ป้,วิยท�7งหมดท�)ได%ร�บการร�กษาแล%วิเห-น ป้ระส�ทธี�ผู้ล 1 คน ซึ่3)ง NNT สามารถช1วิยให%
แพทย�สามารถป้ระเม�น Relative Risk และBenefit ข้องการร�กษาได%
NNT = ARR
1
• NNT = ; ARR = EER-CER
CEREER
1
140
• จ'านวินผู้+%ป้,วิยท�7งหมดท�)ได%ร�บการร�กษาแล%วิได%ผู้ลท�) ไม1พ3งป้ระสงค�ตามมา 1 คน
NNH = ARI
1
• NNH = ; ARI = CER - EER EERCER
1
141
ผู้ลแที่รกซึ่&อน ผู้ลการร)กษา การ
ร)กษา A การ
ร)กษาB
RRR =(CER-EER)/CER
ARR =CER-EER
NNT =1/ARR
ผู้ลที่��เก#ดจร#ง
8%
4%
(8-4)/8 =50%
8-4 =4%
1/4= 25
เมื่��อผู้ลด�มื่าก
80%
40%
(80-40)/80 =50%
80-40 =40%
1/40=2.5
=3 เมื่��อผู้ลไมื่ด�
0.08%
0.04%
(0.08-0.04)/0.08
=50%
0.08-0.04 =0.04%
1/0.04 =2500
CER = Control Event Rate, EER = Experiment Event Rate
142
Random error/Chance,
BiasConfounder
Suthee Rattanamomgkolgul
143
เราอาจ สร�ปการศ6กษาผู้#ด ได&
สร�ปการศ6กษาความื่ส)มื่พ)นธ%ผู้#ด ได&จาก 3 สาเหติ�
•I. ความื่บุ)งเอ#ญ่CHANCE/Random error
•II. อคติ# BIAS
•III. ป9จจ)ยที่�� 3 หร�อ ติ)วกวน-
3rd VARIABLES :
Confounder
144
•การว)ดติ�างๆในการว#จ)ยเก#ดจากการส��มื่ติ)วอย�าง
• จ-านวนติ)วอย�าง มื่�ความื่ส-าค)ญ่ติ�อความื่ ถ1กติ&อง ของการประมื่าณ์ค�าของค�าที่��
แที่&จร#งในธรรมื่ชีาติ# ก)บุค�าที่��ได&จากการว#จ)ย
•Number of sample == Validity
I. ความื่บุ)งเอ#ญ่ Chance
145
ค1าเฉล�)ยข้องน'7าหน�กน�ส�ตเม0)อใช%จ'านวินท�)น'า มาเฉล�)ยต1าง ๆ ก�น
146
N=100
N=25
N=11
147
ผู้ลการร�กษาเบาหวิานด%วิย ยาลดน'7าตาลในเล0อด
เม0)อทดลองด%วิยข้นาดต�วิอย1างท�)แตกต1างก�น
148
Precision of the effect size
Size of the problems
• Prevalence
• Incidence
• Mean
Precision of the size• Prevalence of HT in this
population = 30% (95% CI = 25 - 36%)
• Incidence of skin cancer =100 ติ�อแสน (95% CI= 89-
112 ติ�อแสน)• Mean fasting blood glucose
of the population = 126 mg/mL (95% CI = 80 -
160 %)
149
Confidence interval of the estimates• Prevalence of HT in this
population = 30% (95% CI = 25 - 36%
• ได&มื่าจากการศ6กษาความื่ด)นจากการส��มื่คน1000 คน มื่าจากชี�มื่ชีนแห�งหน6�ง
• หากที่-าการส��มื่ใหมื่� เล�อกคน 1000 คนมื่าว)ด ความื่ด)นอ�กเป3นจ-านวน 100 คร)!ง ค�า
Prevalence ที่��ได&จะมื่� 95 คร)!งที่��ติกอย1�ในชี�วง 25 - 36 %
• มื่�ความื่เชี��อมื่)�นถ6ง 95% ว�า Prevalence of HT ในประชีากรน�!จะอย1�ระหว�าง 25 - 36%
150
II. อคติ# (bias)- มื่� 2 ชีน#ด1. การสร�ปผู้#ดที่��เก#ดจาก การค)ดเล�อกกล��มื่ที่��มื่า
เปร�ยบุเที่�ยบุที่��ไมื่�ย�ติ#ธรรมื่ (selection bias)
2. หร�อการว)ดค�าของ exposure หร�อoutcome ที่��ไมื่�ย�ติ#ธรรมื่ (information bias หร�อ measurement bias)
ที่-าให&เก#ดค�าของความื่ส)มื่พ)นธ% Relative Risk, Odds Ratio, Prevalence Rate Ratio ที่��ผู้#ดไปจากความื่เป3นจร#งในธรรมื่ชีาติ#
151
ควิามแตกต1างระหวิ1าง chance และ bias
Error
Study size
Source: Rothman, 2002
Systematic error (bias)
Random error (chance)
152
Selection bias•ความื่ผู้#ดพลาดที่��เก#ดจากข)!นติอนการค)ดเล�อก
ประชีากรที่��มื่าศ6กษาที่��ไมื่�เหมื่าะสมื่
•มื่�การค)ดเล�อกประชีากรที่��เข&ามื่าในการศ6กษาที่��ไมื่� เที่�าเที่�ยมื่ก)นของกล��มื่ที่��น-ามื่าเปร�ยบุเที่�ยบุที่)!งสอง
ซึ่6�งการเล�อกน)!นไปเก��ยวข&องก)บุ
– สถานะของการเป3น case การเป3น control
– สถานะของการมื่� หร�อไมื่�มื่� ป9จจ)ยที่��ติ&องการศ6กษา
153
Healthy worker effect
154
Healthy worker effect
155
Information Bias
หร�อ measurement bias ความื่ผู้#ดพลาดที่��เก#ดจากข)!นติอนการว)ดค�าของexposure หร�อ outcome ที่��ไมื่�ย�ติ#ธรรมื่
สาเหติ�อาจเก#ดจากผู้1&ว#จ)ย อาสาสมื่)คร หร�อเคร��องมื่�อว)ด
• Subject bias– Recall bias– Public awareness
• Observer bias– Interviewer bias– Follow-up bias
• Diagnostic bias
156
ที่ารกพ#การแติ�ก-าเน#ด VS การได&ร)บุร)งส� X-ray ขณ์ะติ)!ง
ครรภ%• ส)มื่ภาษณ์%ประว)ติ#ของการได&ร)บุร)งส� X-ray
ขณ์ะติ)!งครรภ%•มื่ารดาที่��มื่�บุ�ติรพ#การแติ�ก-าเน#ดจะพยายามื่
อย�างย#�งในการจดจ-ารายละเอ�ยดของสาเหติ� ติ�างๆ ที่��ที่-าให&เก#ดความื่พ#การรวมื่ถ6งการได&ร)บุ
•ในขณ์ะที่��มื่ารดาที่��มื่�บุ�ติรปกติ#แข4งแรงด�น)!นไมื่� ได&พยายามื่จดจ-าหร�อติอบุการให&ส)มื่ภาษณ์%ใน
รายละเอ�ยดที่��เที่�าเที่�ยมื่ก)น
157Unrealistic Odds Ratio (Recall bias)
CaseControl
ExposedNot
Exposed
A B
C D
Hx of radiation
No Hx of radiation
Malformation+No Malformation
OR=AxD BxC
158
Effect of bias
159
Non-differential Misclassification
Case Control
Exposed 200 100
Non-exposed 100 200
Sensitivity 100% , Specificity 60%
Case Control
Exposed 240 60
Non-exposed 180 120
True OR = 4.0 Apparent OR = 2.7- 40 false+
+ 40 false+
- 80 false+
+ 80 false+
Reduction!
160
Differential Misclassification
Case Control
Exposed 200 100
Non-exposed 100 200
Case Control
Exposed 240 60
Non-exposed 100 200
True OR = 4.0 Apparent OR = 8.0
Sensitivity Specificity
Exposed 100% 60%
Non-exposed 100% 100%
161
Misclassification
Measurement error leads to assigning wrong exposure or outcome category
Non-differential Due to random error
Unrelated to exposure or outcome status
Not a bias
Weakens measure of association
Differential Due to systematic error
Related to exposure or outcome status
Bias
Measure of association distorted in any direction
162
Confounder
163
ในป้ระเทศญี่�)ป้�,น ม�การรายงานวิ1าข้นาดข้องรองเท%า (shoe size) เด-กน�กเร�ยนม�ควิามส�มพ�นธี�ก�บ คะแนนข้องการทดสอบควิาม
สามารถทางคณ�ตศาสตร� (Maths Score)
164
When the data were subdivided by age they looked like this
i.e. within each age group there was no association. In this case age was a confounding variable.
165
ConfoundingConfounding
• Not the investigator’s fault, just a fact of life
• Confounding is a mixing of effects between the association of the disease and third factor (the confounder)
• What confounding factors could explain the finding?
166
III. Confounder• จากการว#จ)ยพบุความื่ส)มื่พ)นธ%ที่��เก#ดข6!นจร#ง ๆ• ซึ่6�งอาจเก#ดจากป9จจ)ยที่�� 3 ที่��เก��ยวข&องก)บุ ป9จจ)ยและการ
เก#ดโรคที่��ก-าล)งศ6กษา• การด��มื่ Alcohol VS การเก#ดมื่ะเร4งปอด• ป9จจ)ยที่�� 3 การส1บุบุ�หร��• ป9จจ)ยที่�� 3 ที่��มื่�ผู้ลติ�อความื่ส)มื่พ)นธ%ที่��ส)งเกติเห4น เป3นไปได&
2 ประการ ; - Confounder และ Effect
modifier or Interaction
167
Alcohol VS Lung Cancer
CA Lung
+
CA Lung-
Alcohol+
800 200
Alcohol-
200 800
1000 1000OR = 800x800 = 4 200x200
OR=AxD BxC
168
Alcohol VS CA Lung
•Odds Ratio = 4
• ผู้1&ที่��ด��มื่ Alcohol มื่�โอกาสเป3นมื่ะเร4งปอดมื่ากกว�าผู้1& ที่��ไมื่�ด��มื่ 4 เที่�า
• ผู้ลที่��ได&น�!มื่าจากการศ6กษาที่��ไมื่�มื่� Bias
•เป3นปรากฏการณ์%ที่��เก#ดข6!นจร#ง
•แติ�มื่)นจะเหมื่าะสมื่ที่��จะรายงานผู้ลการศ6กษาแบุบุน�!หร�อไมื่�?
169
Alcohol VS CA Lung• มื่�ความื่เป3นไปได&ไหมื่ที่��ความื่ส)มื่พ)นธ%ที่��เก#ดข6!นน)!นจร#ง
ๆ แล&วเก#ดจากป9จจ)ยอ��น
• ในความื่เป3นจร#งในธรรมื่ชีาติ# Alcohol อาจไมื่�ได&เป3นติ&นเหติ�ของการเก#ดมื่ะเร4งปอดก4ได&
• Alcohol อาจเป3นเพ�ยงเหติ�การณ์%ที่��เก#ดข6!นมื่า พร&อมื่ ๆ ก)น
จากความื่ส)มื่พ)นธ%ที่��แที่&จร#งของป9จจ)ยอ��น• มื่าค#ดด1 ถ6งผู้ลของการส1บุบุ�หร�� ก)บุความื่ส)มื่พ)นธ%ค1�น�!
ระหว�าง Alcohol ก)บุ มื่ะเร4งปอด
170
CA Lung
+
CA Lung
-Alcohol+ 800 200
Alcohol- 200 800
1000 1000
OR = 800x800 = 4 200x200
CA Lung
+
CA Lung
-Alcohol+ 200 25
Alcohol- 600 75
800 100
OR = 1
SMOKE NON – SMOKE
OR = 1
CA Lung
+
CA Lung
-Alcohol+ 50 225
Alcohol- 150 675
200 900
171
Confounder
SMOKING
CA LungAlcohol
SMOKING
CA LungAlcohol
(1) (2)
172
การจ)ดการก)บุConfounder
• Restriction inclusion criteria
• Matching (in case-control study) by Smoking status or by Gender
• Stratified analysis Mantel Haenszel Method
• Multivariate analysis • ในการศ6กษาแบุบุ RCT ใชี&ว#ธ#การ
randomisation
173
Confounder
CD+4 Level
Time to AIDS DeathHIV-1 RNA
•Not be an intermediate of the association between the exposure and the outcome
(3)
174
Confounding is
• An incidental factor can be a confounder if it satisfies the following conditions:(1) Must be associated with exposure(2) Must be an independent risk factor for the outcome(3) Must not be an intermediate step in the causal chain
between exposure and outcome(4) Must be present to a greater or lesser degree in the
study group vs. the comparison group• If confounder is unequally distributed between study and
comparison groups → Confounding• If confounder is equally distributed between study and
comparison groups → NO Confounding (even if the 1st three conditions are met)
175
Interpreting Associations Between Exposure and Disease
2. Is it due to bias?
3. Is it due to confounding?
Is it causal? According to causal criteria
1. Is it the result of chance?
Intro
176
External Validity
• Generalizability• Degree to which study results are relevant for
populations other than the target population
• Must use judgment as to whether the findings would make sense in other populations– No simple formula for determining external validity
177
Screening test
เคร0)องม0อตรวิจค�ดกรอง
178
Clinical Diagnosis
Disease Onset
No Disease Asymptomatic Disease
Clinical Course
Primary
Remove Risk Factors
Secondary
Early Detection and Treatment
Tertiary
Reduce Complications
Levels of Prevention
Fletcher RF, Fletcher SW, Wagner EH. Clinical Epidemiology: The Essentials, 3 rd ed. Williams and Wilkins, Baltimore, 1996.
179
Sojourn time and Lead time
No detectable disease
t0 t2 t1
Clinically detectable (symptomatic) disease
Period in which cancer is screen detectable
Sojourn time
Lead time
Time
Screen time
180
Ideal Screening Program:
Features of the diseaseSignificant impact on public health
Asymptomatic phase during which detection is possible
Outcomes improved by treatment during asymptomatic phase
Goroll AH, May LA, Mulley AG. Primary Care Medicine. 3d ed. Philadelphia: Lippincott, 1995: 13-6.
181
Ideal Screening Program:
Features of the test
Sufficiently sensitive to detect disease during asymptomatic phase
Sufficiently specific to minimize false positive results
Acceptable to patients
Goroll AH, May LA, Mulley AG. Primary Care Medicine. 3d ed. Philadelphia: Lippincott, 1995: 13-6.
182
Ideal Screening Program:
Features of the screened populationSufficiently high prevalence of the disease to
justify screening
Relevant medical care is accessible
Patients willing to comply with further work-up and treatment
Goroll AH, May LA, Mulley AG. Primary Care Medicine. 3d ed. Philadelphia: Lippincott, 1995: 13-6.
183
Disease Present Disease Absent
RE SULT
+
True PositiveWill benefit from test, (have disease that will be revealed).(A)
False PositiveRequire add’l tests, procedures to prove no ds.(B)
Positive predictive value:% with + test who have ds.A/(A+B)
_
False NegativeFalsely reassured they do not have ds.(C)
True NegativeFound to have no ds. but may have had anxiety about test (D)
Negative predictive value:% with – test with no ds. D/(C+D)
Sensitivity:% with ds. with + testA/(A+C)
Specificity:% without ds. with - test D/(B+D)
Operating Characteristics
184
Bias in the Evaluation of Screening Tests
“Early diagnosis will always appear to improve survival, even when the therapy is worthless.”
Sackett DL, Haynes RB, et al. Early Diagnosis. In: Clinical Epidemiology: a basic science for clinical medicine. 2d ed. Boston: Little, Brown, 1991: 153-170.
185
Compliance/Volunteer Bias
Type of selection bias
Volunteers are healthier, with lower mortality from all causes
Observed benefit may be due to self-selection of volunteers
186
Overdiagnosis Bias
Detection of disease that would never have become clinically evident during
the time of the trial and follow-up
187
Clinical illness
2 years
Lead-Time Bias
4 yearsDeath
Diagnosis at Clinical
Presentation
3 years
Natural History
Unscreened Population
Screened Population
Clinical symptoms
Diagnosis at Clinical
Presentation
Death
3 years
Gates TJ, Screening for Cancer: Evaluating the Evidence. Am Fam Physician. 2001;63:513-22
Death3 years
Screening Diagnosis5 years
Death
Asymptomatic periodOnset
188
Length-time Bias
Aggressive Disease
Onset Clinical
Presentation
Death
Clinical
PresentationDeathOnset
1 yr sx
Screening interval
1 year
6 mo. asymptomatic period
2 year asymptomatic period
4 yr sx
Less Aggressive Disease
Gates TJ, Screening for Cancer: Evaluating the Evidence. Am Fam Physician. 2001;63:513-22.
189
190
1. Screen for Risk factors การค�ดกรองป้9จจ�ยเส�)ยง - เพ0)อค%นหากล�1มคนท�)ม�ป้9จจ�ยเส�)ยงต1อการเก�ดโรคน
ช�มชน เช1น การค�ดกรองผู้+%ท�)ม�ระด�บไข้ม�นในเล0อดส+ง การใช%ป้ระวิ�ต�ในการค�ดกรองหญี่�งท�)ม�ป้9จจ�ยเส�)ยง
ต1อควิามด�นโลห�ตส+ง เช1น ส+บบ�หร�)จ�ด ม� ควิามเคร�ยดส+ง ม�ป้ระวิ�ต�คนในครอบคร�วิ
2. Screen for Diseases การค�ดกรองโรค - เพ0)อค%นหาโรคในกล�1มป้ระชากรและหากสงส�ยวิ1าจะม�
โรคจะท'าการตรวิจวิ�น�จฉ�ยต1อไป้ เช1น การใช% pap smear ค�ดกรองโรคมะเร-งป้าก
มดล+ก
Objectives of screening in community
191
IN EVERY 100 PEOPLE, 4 WILL HAVE THE DISEASE
Disease +
4
Disease -
96
Population
100
If these 100 people are representative of the population at risk, the assessed rate of those with the
disease (4%) represents the PREVALENCE of the disease – it can also be considered the PRE-TEST
PROBABILITY of having the disease
192
OF THE 4 PEOPLE WITH THE DISEASE, THE TEST WILL DETECT 3
Disease +
4
Disease -
96
Test +
3
Test -
1
Population
100
In other words, the sensitivity is
75%
193
AMONG THE 96 PEOPLE WITHOUT THE DISEASE, 7 WILL TEST POSITIVE
Disease +
4
Disease -
96
Test +
7
Test -
89
Test +
3
Test -
1
Population
100
In other words, the specificity is
93%
194
POSITIVEPREDICTIVE
VALUE = 30%
AMONG THOSE WHO TEST POSITIVE, 3 IN 10 WILL ACTUALLY HAVE THE DISEASE
Disease +
4
Disease -
96
Test +
7
Test -
89
Test +
3
Test -
1
Population
100
This is also the POST-TEST PROB- ABILITY of having
the disease
195
NEGATIVEPREDICTIVEVALUE = 99%
AMONG THOSE WHO TEST NEGATIVE, 89 OF 90 WILL NOT HAVE THE DISEASE
Disease +
4
Disease -
96
Test +
7
Test -
89
Test +
3
Test -
1
Population
100
196
CONVERSELY, IF SOMEONE TESTS NEGATIVE, THE CHANCE OF HAVING THE
DISEASE IS ONLY 1 IN 90
Disease +
4
Disease -
96
Test +
7
Test -
89
Test +
3
Test -
1
Population
100
197
PREDICTIVE VALUES AND CHANGING PREVALENCE
Disease +
4
Disease -
996
Population
1000
Prevalence reduced by an order of magnitude from 4%
to 0.4%
198
PREDICTIVE VALUE AND CHANGING PREVALENCE
Disease +
4
Disease -
996
Test +
70
Test -
926
Test +
3
Test -
1
Population
1000Sensitivity
and Specificity unchanged
199
POSITIVEPREDICTIVEVALUE = 4%
POSITIVE PREDICTIVE VALUE AT LOW PREVALENCE
Disease +
4
Disease -
996
Test +
70
Test -
926
Test +
3
Test -
1
Population
1000Previously,
PPV was 30%
200
NEGATIVEPREDICTIVEVALUE >99%
NEGATIVE PREDICTIVE VALUE AT LOW PREVALENCE
Disease +
4
Disease -
996
Test +
70
Test -
926
Test +
3
Test -
1
Population
1000
Previously, NPV was 99%
201
PREDICTION OF LOW PREVALENCE EVENTS
• Even highly specific tests, when applied to low prevalence events, yield a high number of false positive results
• Because of this, under such circumstances, the Positive Predictive Value of a test is low
• However, this has much less influence on the Negative Predictive Value
202
RELATIONSHIP BETWEEN PREVALENCE AND PREDICTIVE VALUE
0
0.2
0.4
0.6
0.8
1
0.05 0.2 0.4 0.6 0.8 0.95
Pre-test Probability (Prevalence)
Pre
dic
tive V
alu
e
PPVNPV
Based on a test with 90% sensitivity and 82% specificity
Difference between PPV and
NPV relatively small
Difference between PPV and
NPV relatively large
203
Diagnostic tests:Likelihood ratios
Group Test
Not-normal
Normal
x > +
a b a+b
x -
c d c+d
a+c b+d n
• Positive likelihood ratio [a/(a+c)] / [b/(b+d)]
• Negative likelihood ratio [c/(a+c)] / [d/(b+d)]
They do not depend as much on the prevalence !!
Shrikant I. Bangdiwala, Ph.D. UNC
204
Positive Likelihood Ratio (LR+)
• Reflects the degree of confidence that a person who scores in the positive does have the disease– sensitivity / (1-specificity)
• The higher the LR+, the more informative is the test for identifying people who have the disorder
Ref: Sackett 2000 Shrikant I. Bangdiwala, Ph.D. UNC
205
Interpreting LR+ values (Sackett et al., 1991)
• LR+ > 20 Very high– Virtually certain that a person with this score has
the disorder
• LR+ = 10 High– Disorder very likely in a person with this score
• LR+ = 4 Intermediate– Test is suggestive of disorder but insufficient to
diagnose
• LR+ = 1 Equivocal– A person who scores in the disordered range on the
measure may or may not have the disorder; the measure provides no new informationRef: Sackett 2000
Shrikant I. Bangdiwala, Ph.D. UNC
206
Negative Likelihood Ratio (LR-)
• Reflects the degree of confidence that a person scoring in the negative (normal) range on the diagnostic indicator truly does not have the disorder – (1-sensitivity) / specificity
• The lower the LR-, the more informative the test is for ruling out the presence of disorder
Ref: Sackett 2000
Shrikant I. Bangdiwala, Ph.D. UNC
207
Interpreting LR- values (Sackett et al., 1991)
• LR- < 0.10 Very low– virtually certain that a person scoring in this range
does not have the disorder
• LR- = 0.20 Low– disorder very unlikely
• LR- = 0.40 Intermediate– test is suggestive but insufficient to rule out the
disorder
• LR- = 1.0 Equivocal– A person scoring in the normal range on this measure
may or may not be normalRef: Sackett 2000
Shrikant I. Bangdiwala, Ph.D. UNC
208
80 a
30 b
c 20
d 70
Disorder Status (re: Gold Standard)
+ Disorder (LI) - Disorder (LN)
New Test Result
+ Disorder (LI)
-Disorder (LN)
100100 100100Sens = .80Spec = .70LR+ = sens/(1-spec) = .80/.30 = 2.67LR- = (1-sens)/spec = .20/.770 = 0.29
Shrikant I. Bangdiwala, Ph.D. UNC
209
METHOD 4: A TEST WITH NORMALLY DISTRIBUTED VALUES
Negative Positive
Degree of ‘positivity’ on test
% o
f G
rou
p
DISEASED
NON-DESEASED
Test cut-off
Assessing the performance of the test assumes that these two
distributions remain constant. However, each
of them will vary (particularly through spectrum or selection
bias)
210
CASESNON-CASES
PERFORMANCE OF A DIAGNOSTIC TEST
Negative Positive
Degree of ‘positivity’ on test
% o
f G
rou
p
DISEASED
NON-DESEASED
Test cut-offFALSE
NEGATIVESFALSE
POSITIVES
211
MINIMISING FALSE NEGATIVES: A SENSITIVE TEST
Negative Positive
Degree of ‘positivity’ on test
% o
f G
rou
p
DISEASED
NON-DESEASED
Test cut-off
Cut-off shifted to minimise false negatives ie to optimise sensitivity
CONSEQUENCES:
- Specificity reduced
- A Negative result from a seNsitive test rules out the diagnosis - snNout
CASESNON-CASES
212
MINIMISING FALSE POSITIVES: A SPECIFIC TEST
Negative Positive
Degree of ‘positivity’ on test
% o
f G
rou
p
DISEASED
NON-DESEASED
Test cut-off
Cut-off shifted to minimise false positives ie to optimise specificity
CONSEQUENCES:
- Sensitivity reduced
- A Positive result from a sPecific test rules in the diagnosis - spPin
213
Trade- Off between Sensitivity and Specificity when Diagnosing Diabetes
Blood Sugar Level 2 hr after Eating
(mg/100mL)
Sensitivity (%)
Specificity (%)
70
98.6
8.8
80
97.1
25.5
90
94.3
47.6
100
88.6
69.8
110
85.7
84.1
120
71.4
92.5
130
64.3
96.9
140
57.1
99.4
150
50.0
99.6
160
47.1
99.8
170
42.9
100.0
180
38.6
100.0
190
34.3
100.0
200
27.1
100.0
214
RECEIVER OPERATING CHARACTERISTIC CURVE
Overall shape is predicted by the reciprocal relationship between sensitivity and specificity
The closer the curve gets to Sensitivity=1 and Specificity=1, the better the overall performance of the test
The diagonal line (representing Sensitivity=0.5 and Specificity=0.5) represents performance no better than chance
Hence the area under the curve gives a measure of the test’s performance
FALSE POSITIVE RATE (1-Specificity)
TR
UE P
OS
ITIV
E R
ATE
(Sen
sit
ivit
y)
215
0
100
1-Specificity
Sensi
tivi
tyAREA UNDER ROC CURVES
0
100
1-Specificity
Sensi
tivi
ty Sensitivity and specificity both 100% - TEST PERFECT
Sensitivity and specificity both 50% - TEST USELESS
AREA=1.0
AREA=0.5The area under a ROC curve will be between 0.5 and 1.0
216
0
100
1-Specificity
Sensi
tivi
tyAREA UNDER ROC CURVES
Area = 0.7 (between 0.5 and
1.0)
•Consider (hypothetically) two patients drawn randomly from the DISEASE+ and DISEASE- groups respectively
• If the test is used to guess which patient is from the DISEASE+ group, it will be right 70% of the time
217
RECEIVER OPERATING CHARACTERISTIC (ROC) CURVE
0102030405060708090
100
0 20 40 601-Specificity
Sensi
tivit
y
ACAT
MC
This study compared the performance of a dementia screening test in a community sample (ACAT) and a memory clinic sample (MC)
Flicker L, Loguidice D, Carlin JB, Ames D. The predictive value of dementia screening instruments in clinical populations. International Journal of Geriatric Psychiatry 1997 ; 12 : 203-209
218
Improving Predictive Value
• It is possible to increase the pre-test likelihood (and thereby improve the predictive value) by targeting testing to those at highest risk, or by selectively screening specific populations.
219
220
221
Types of diagnostic tests
• Qualitative ordinal categorical data– E.g. clinical examination and determination of
categories of severity of illness– E.g. psychological assessment of mental state
• Quantitative continuous data– E.g. laboratory measurement of blood
chemistry
Shrikant I. Bangdiwala, Ph.D. UNC
222
Combining tests
• Goal: To increase the confidence in a given decision
• Methods:– Parallel tests– Tests in series– Screening/
confirmatory tests– Re-testing negatives
Shrikant I. Bangdiwala, Ph.D. UNC
223
Parallel testing
• Decision: positive if any of the tests is positive
• Preferred if starting treatment quickly is valuable
• Effects:– Increases sensitivity and
negative predictive value less false negatives
– Decreases specificity and positive predictive value more false positives
Shrikant I. Bangdiwala, Ph.D. UNC
224
Tests in series
• Decision: positive if all of the tests are positive
• Preferred if treatment may have side effects
• Effects:– Increases specificity
and positive predictive value less false positives
– Decreases sensitivity and negative predictive value more false negatives
Shrikant I. Bangdiwala, Ph.D. UNC
225
Screening / confirmatory tests
Special case of several tests with following characteristics:
• Test 1: – High sensitivity– Reasonable specificity– Low cost
• Test 2:– Highly specific
Shrikant I. Bangdiwala, Ph.D. UNC
226
Re-testing negativesSpecial case of several
tests with following characteristics:– Same test is applied
repeatedly at different time points
– Those already positive are not re-tested
• Effects: – Increases sensitivity at
the aggregate level
– Specificity becomes more important with decreasing prevalence
TIME 1 TIME 2 TIME 3
Shrikant I. Bangdiwala, Ph.D. UNC
227
Studying prognostic factor
228
Stage of Pre-pathologic
onset
Pre-symptomatic stage
Clinically manifest disease
Progress to a fatal termination
Remission and relapses
Regress spontaneously, leading to recovery
Risk Factors Effect of Treatment Prognostic factor
Natural history of disease
Screening
229
Survival Curves
1 year survival 95%Median survival unknown
1 year survival 20%Median survival 3 months
1 year survival 20%Median survival 9 months
1 year survival 20%Median survival 7 months
230
Survival curves
• A plot of outcome over time for each group
• At predetermined intervals, measure how many people have the outcome compared to how many were at risk at the beginning of the period
• Can compare the curves for two different groups and see if they are different
231
The differential blood pressure sign in general practice: prevalence and prognostic value.
Clark CE, Powell RJ. Fam Pract. 2002 Oct;19(5):439-41.
232
233
234
235
Kaplan Meier Method
Time ti
(month)
No. at risk No. events Survival rate
0 20 0 1.00
5 20 2 [1-(2/20)]*1.00=0.90
6 18 0 [1-(0/18)]*0.90=0.90
10 15 1 [1-(1/15)]*0.90=0.84
13 14 2 (1-(2/14)]*0.84=0.72
Follow up 20 patients with lung cancer for 13 months
236
Kaplan Meier Curve
0.60.7
0.80.9
1.00 5 10 15 20
Survival Time
0.6
0.7
0.8
0.9
1.0
Pro
port
ion
Sur
vivi
ng (
95%
Con
fiden
ce)
237
Kaplan Meier
• One way to estimate survival
• Nice, simple, can compute by hand
• Can add stratification factors
• Cannot evaluate covariates like Cox model
• No sensible interpretation for competing risks
238
239
Hazard function (h(t))
• Defn: the probability of an event at a specific moment in time (t), given the patient has already survived to that point in time.
– closely linked to the survival function. – indicates the probability of the patient "failing"
during the next time period. – a direct measure of prognosis.
240
Prognostic factors and clinical outcome in acute lower respiratory tract infections: a prospective study in general practice.
• BACKGROUND: Unrealistic expectations about illness duration are likely to result in reconsultations and associated unnecessary antibiotic prescriptions. An evidence-based account of clinical outcomes in patients with lower respiratory tract infection (LRTI) may help avoid unnecessary antibiotic prescriptions and reconsultations.
• OBJECTIVES: We aimed to identify clinical factors that may predict a prolonged clinical course or poor outcome for patients with LRTI and to provide an evidence-based account of duration of an LRTI and the impact of the illness on daily activities in patients consulting in general practice.
• METHODS: A prospective cohort study of 247 adult patients with a clinical diagnosis of LRTI presenting to 25 GPs in The Netherlands was carried out. Multivariable Cox regression analysis was used to identify baseline clinical and infection parameters that predicted the time taken for symptoms to resolve. A Kaplan-Meier curve was used to analyse time-to-symptom resolution. Clinical cure was recorded by the GPs at 28 days after the initial consultation and by the patients at 27 days.
Fam Pract. 2006 Oct;23(5):512-9. Epub 2006 Jun 20.Hopstaken RM, Coenen S, Butler CC, Nelemans P, Muris JW, Rinkens PE, Kester AD, Dinant GJ.Institution of Health Centres Eindhoven Eindhoven, GC Meerhoven, Eindhoven, The Netherlands. [email protected]
241
242
RESULTS: • Co-morbidity of asthma was a statistically significant predictor
of delayed symptom resolution, whereas the presence of fever, perspiring and the prescription of an antibiotic weakly predicted enhanced symptom resolution.
• The GPs considered 89% of the patients clinically cured at 28 days, but 43% of these nevertheless reported ongoing symptoms.
• Patient-reported cure was much lower (51%), and usual daily activities were limited in 73% of the patients at baseline, and 19% at final follow-up.
CONCLUSIONS: • The course of LRTI was generally uncomplicated, but the
morbidity of this illness was considerable with a longer duration than generally reported, especially for patients with co-existent asthma.
• These results underline once again the importance of providing GPs with an evidence-based account of outcomes to share with patients in order to set realistic expectations and of enhancing their communication skills within the consultation.
243
244
Hazard function (h(t))
• Defn: the probability of an event at a specific moment in time (t), given the patient has already survived to that point in time.
– closely linked to the survival function. – indicates the probability of the patient "failing"
during the next time period. – a direct measure of prognosis.
245
Prognostic factors and clinical outcome in acute lower respiratory tract infections: a prospective study in general practice.
• BACKGROUND: Unrealistic expectations about illness duration are likely to result in reconsultations and associated unnecessary antibiotic prescriptions. An evidence-based account of clinical outcomes in patients with lower respiratory tract infection (LRTI) may help avoid unnecessary antibiotic prescriptions and reconsultations.
• OBJECTIVES: We aimed to identify clinical factors that may predict a prolonged clinical course or poor outcome for patients with LRTI and to provide an evidence-based account of duration of an LRTI and the impact of the illness on daily activities in patients consulting in general practice.
• METHODS: A prospective cohort study of 247 adult patients with a clinical diagnosis of LRTI presenting to 25 GPs in The Netherlands was carried out. Multivariable Cox regression analysis was used to identify baseline clinical and infection parameters that predicted the time taken for symptoms to resolve. A Kaplan-Meier curve was used to analyse time-to-symptom resolution. Clinical cure was recorded by the GPs at 28 days after the initial consultation and by the patients at 27 days.
Fam Pract. 2006 Oct;23(5):512-9. Epub 2006 Jun 20.Hopstaken RM, Coenen S, Butler CC, Nelemans P, Muris JW, Rinkens PE, Kester AD, Dinant GJ.Institution of Health Centres Eindhoven Eindhoven, GC Meerhoven, Eindhoven, The Netherlands. [email protected]
246
สร�ป้การวิ�ดทางระบาดวิ�ทยา•ควิามเส�)ยงข้องการเก�ดโรค Incidence rate
•ข้นาดข้องป้9ญี่หา Prevalence rate
•ควิามส�มพ�นธี�ข้องโรคก�บป้9จจ�ยเส�)ยง
Relative risk (RR)Incidence rate ratioOdds ratio (OR)
•ผู้ลกระทบข้องป้9จจ�ยเส�)ยงต1อการเก�ดโรคในป้ระชากร
Population attributable risk (PAR)Population prevented fraction
•ผู้ลกระทบข้องการร�กษา Number needed to treat (NNT)Number needed to harm (NNH)
• ต%องการเป้ร�ยบเท�ยบincidence/mortality ระหวิ1างป้ระชากร
Direct and indirect standardization Standardized mortality ratio (SMR)
247
Thank you for your kind attentions
248
วิ�ทยาการระบาดในงานเวิชศาสตร�ครอบคร�วิ
นพ. ส�ธี�ร� ร�ตนะมงคลก�ลภาควิ�ชาเวิชศาสตร�ป้�องก�นและส�งคม
มหาวิ�ทยาล�ยศร�นคร�นทรวิ�โรฒ