Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given...
Transcript of Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given...
![Page 2: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/2.jpg)
Objective: To quantify the magnitude and direction of systematic error (bias)
Quantitative bias analysis
![Page 3: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/3.jpg)
Systematic error Random error
![Page 4: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/4.jpg)
Quantify bias from systematic error
- Confounding
- Selection bias
- Misclassification
Quantitative bias analysis
![Page 5: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/5.jpg)
Quantify bias from systematic error
- Confounding
- Selection bias
- Misclassification
Quantitative bias analysis
![Page 6: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/6.jpg)
Reviewer #2
![Page 7: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/7.jpg)
‘The authors failed to take into account confounding by _ _ _ _ _ _ _ _ _
![Page 8: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/8.jpg)
Smoking
Alcohol
Frailty
BMI
Education
Exercise
![Page 9: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/9.jpg)
1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able to adjust for this confounder?
Quantitative bias analysis
Point estimate with bias →Point estimate taking into account bias
![Page 10: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/10.jpg)
1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able to adjust for this confounder?
Quantitative bias analysis
![Page 11: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/11.jpg)
1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able to adjust for this confounder?
2. How strong would an unknown confounder have to be to fully explain the observed effect?
Quantitative bias analysis
![Page 12: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/12.jpg)
Quantitative bias analysis
1. Simple bias analysis
2. Probabilistic bias analysis
![Page 13: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/13.jpg)
Quantitative bias analysis
1. Simple bias analysis
2. Probabilistic bias analysis
![Page 14: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/14.jpg)
1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able to adjust for this confounder?
2. How strong would an unknown confounder have to be to fully explain the observed effect?
![Page 15: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/15.jpg)
1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able to adjust for this confounder?
2. How strong would an unknown confounder have to be to fully explain the observed effect?
![Page 16: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/16.jpg)
PC1: Prevalence of confounder among exposedPC0: Prevalence of confounder among unexposedRRCD: Relative risk of confounder associated with outcome
![Page 17: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/17.jpg)
PC1: Prevalence of confounder among exposedPC0: Prevalence of confounder among unexposedRRCD: Relative risk of confounder associated with outcome
![Page 18: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/18.jpg)
PC1
PC0
RRCDBias parameters
![Page 19: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/19.jpg)
Example
Does male circumcision protect against HIV infection?
Possible confounder: Being Muslim
RR not adjusted for religion: 0.35 (95% CI 0.28 to 0.44)Tyndall et al., 1996
![Page 20: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/20.jpg)
PC1
PC0
RRCDBias parameters
![Page 21: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/21.jpg)
![Page 22: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/22.jpg)
PC1
PC0
RRCD
0.80.10.65
RR =0.35
0.8 0.65−1 +1
0.1 0.65−1 +1
=0.47
![Page 24: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/24.jpg)
![Page 25: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/25.jpg)
![Page 26: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/26.jpg)
![Page 27: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/27.jpg)
![Page 28: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/28.jpg)
![Page 29: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/29.jpg)
1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able to adjust for this confounder?
2. How strong would an unknown confounder have to be to fully explain the observed effect?
![Page 30: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/30.jpg)
2. How strong would an unknown confounder have to be in order to fully explain the observed effect?
RRCD
RRECBias parameters
![Page 31: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/31.jpg)
Example
How strong would confounding be to explain the effect of smoking on lung cancer?
RR 10.73, 95% CI 8.02 to 14.36
![Page 32: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/32.jpg)
VanderWeele and Ding, 2017
The E-value is the minimum strength of association on the relative risk scale that a confounder would need to have with both the treatment and outcome to explain away the observed association
![Page 34: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/34.jpg)
RR
CD
REEC
Mathur, 2019
![Page 35: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/35.jpg)
![Page 36: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/36.jpg)
E-value for point estimate: 7.4
E-value for lower limit of CI: 6.8
![Page 37: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/37.jpg)
´The observed risk ratio could be explained away by an unmeasured confounder that was associated with both hydrochlorothiazide and skin cancer by a risk ratio of 7.4-fold each.The confidence interval could be moved to the null by an unmeasured confounder associated with both hydrochlorothiazide and skin cancer by a risk-ratio of 6.8-fold’
![Page 38: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/38.jpg)
Quantitative bias analysis
1. Simple bias analysis
2. Probabilistic bias analysis
![Page 39: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/39.jpg)
1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able to adjust for this confounder?
2. How strong would an unknown confounder have to be to fully explain the observed effect?
![Page 40: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/40.jpg)
Bias parameters →
Probability distribution of bias parameters
![Page 41: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/41.jpg)
![Page 42: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/42.jpg)
1. Assign probability distributions to each bias parameter
2. Randomly sample from the bias parameter distributions
3. Use simple bias analysis to correct for the bias
4. Resample, save, and summarize
![Page 43: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/43.jpg)
![Page 44: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/44.jpg)
![Page 45: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/45.jpg)
![Page 46: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/46.jpg)
Example
Does circumcision protect against HIV infection?
Possible confounder: being muslim
RR not adjusted for religion: 0.35 (95% CI 0.28 to 0.44)
![Page 47: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/47.jpg)
Example
1. Assign probability distributions to each bias parameter
2. Randomly sample from the bias parameter distributions
3. Use simple bias analysis to correct for the bias
4. Resample, save, and summarize
![Page 48: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/48.jpg)
Ressoruces for probabilistic bias analysis in Excel and SAS:https://sites.google.com/site/biasanalysis/
![Page 49: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/49.jpg)
![Page 50: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/50.jpg)
![Page 51: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/51.jpg)
Example
1. Assign probability distributions to each bias parameter
2. Randomly sample from the bias parameter distributions
3. Use simple bias analysis to correct for the bias
4. Resample, save, and summarize
![Page 52: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/52.jpg)
Stata
![Page 53: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/53.jpg)
![Page 54: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/54.jpg)
![Page 55: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/55.jpg)
![Page 56: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/56.jpg)
episensi 105 85 527 93, st(cs) reps(10000)
dpunexp(trapezoidal(0.03 0.04 0.07 0.10))
dpexp(trapezoidal(0.7 0.75 0.85 0.9))
drrcd(trapezoidal(0.5 0.6 0.7 0.8))
grarrtot grprior nodot
![Page 57: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/57.jpg)
episensi 105 85 527 93, st(cs) reps(10000)
dpunexp(trapezoidal(0.03 0.04 0.07 0.10))
dpexp(trapezoidal(0.7 0.75 0.85 0.9))
drrcd(trapezoidal(0.5 0.6 0.7 0.8))
grarrtot grprior nodot
![Page 58: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/58.jpg)
episensi 105 85 527 93, st(cs) reps(10000)
dpunexp(trapezoidal(0.03 0.04 0.07 0.10))
dpexp(trapezoidal(0.7 0.75 0.85 0.9))
drrcd(trapezoidal(0.5 0.6 0.7 0.8))
grarrtot grprior nodot
![Page 59: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/59.jpg)
episensi 105 85 527 93, st(cs) reps(10000)
dpunexp(trapezoidal(0.03 0.04 0.07 0.10))
dpexp(trapezoidal(0.7 0.75 0.85 0.9))
drrcd(trapezoidal(0.5 0.6 0.7 0.8))
grarrtot grprior nodot
![Page 60: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/60.jpg)
episensi 105 85 527 93, st(cs) reps(10000)
dpunexp(trapezoidal(0.03 0.04 0.07 0.10))
dpexp(trapezoidal(0.7 0.75 0.85 0.9))
drrcd(trapezoidal(0.5 0.6 0.7 0.8))
grarrtot grprior nodot
![Page 61: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/61.jpg)
episensi 105 85 527 93, st(cs) reps(10000)
dpunexp(trapezoidal(0.03 0.04 0.07 0.10))
dpexp(trapezoidal(0.7 0.75 0.85 0.9))
drrcd(trapezoidal(0.5 0.6 0.7 0.8))
grarrtot grprior nodot
![Page 62: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/62.jpg)
![Page 63: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/63.jpg)
![Page 64: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/64.jpg)
![Page 65: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/65.jpg)
episensi 105 85 527 93, st(cc) reps(1000)
dpunexp(trapezoidal(0.03 0.04 0.07 0.10))
dpexp(trapezoidal(0.7 0.75 0.85 0.9))
drrcd(log-n(-0.43 0.11))
grarrtot grprior nodot
![Page 66: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/66.jpg)
![Page 67: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/67.jpg)
![Page 68: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/68.jpg)
![Page 69: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/69.jpg)
Quantitative bias analysis
1. Simple bias analysis2. Probabilistic bias analysis
- Confounding- Selection bias- Misclassification
![Page 70: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/70.jpg)
Probabilistic bias analysis
- Confounding
- Selection bias
- Misclassification
Accounting for several biases at the same time: Multiple bias modeling
![Page 73: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/73.jpg)
![Page 74: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/74.jpg)
![Page 75: Quantitative bias analysis - Dansk Selskab for Farmakoepidemiologi · 2019-05-03 · 1. Given confounding by an unmeasured variable, what effect estimate would remain if we were able](https://reader033.fdocuments.us/reader033/viewer/2022060305/5f0945ec7e708231d42608c5/html5/thumbnails/75.jpg)