+ All Categories
Home > Documents > Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA...

Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA...

Date post: 04-Jul-2020
Category:
Upload: others
View: 4 times
Download: 0 times
Share this document with a friend
23
Adaptive designs for diagnostic accuracy studies
Transcript
Page 1: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

Adaptive designs fordiagnostic accuracy studies

Page 2: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

07/11/2017 Adaptive designs for DTA studies 2

Although adaptive designs cannot ‘change theanswer’ regarding the accuracy of a particulardiagnostic test, they can increase the efficiency infinding an answer.

[adopted from Kairalla et al., 2012]

Page 3: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

07/11/2017 Adaptive designs for DTA studies 3

Adaptive study designs

Diagnosticaccuracy studies

Blindedinterim analysis

Adaptationsregarding:‐ prevalence‐ % discordant results‐ % missing values‐ reference standard‐ external: noninferiority margins,cutoff value

Unblindedinterim analysis

Adaptationsregarding:‐ estimated accuracy‐ target population‐ comparator‐ hypotheses

Randomizeddiagnostic studies

Blindedinterim analysis

Adaptationsregarding:diagnosticaccuracy orbenefit‐riskratio(external)

Unblindedinterim analysis

Adaptationsregarding:‐ proportion of TP, TN,FP, FN or benefit‐riskratio (internal)‐ study design

Adaptiveseamless designs

Page 4: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

07/11/2017 Adaptive designs for DTA studies 4

[FDA, 2016]

Page 5: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

Structure

• Blinded interim analyses

• Unblinded interim analysis

• To look ahead

07/11/2017 Adaptive designs for DTA studies 5

Page 6: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

Structure

• Blinded interim analyses

• Unblinded interim analysis

• To look ahead

07/11/2017 Adaptive designs for DTA studies 6

Page 7: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

07/11/2017 Adaptive designs for DTA studies 7

Adaptations regarding:‐ prevalence‐ % discordant results‐ % missing values‐ reference standard‐ disease spectrum‐ accuracy of the comparator?

[FDA, 2016]

Modifications based entirely on information from a source completely external to the study [e.g. cutoff value, non‐inferiority margins] are not adaptive designs...

Page 8: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

• The lower the conditional dependence between the tests,

the larger the sample size

→ largest sample size for maximum negative dependence

• Sensitivity: 

• Specificity:

/

= prevalence

Sample size re‐estimation in pairedcomparative diagnostic accuracy studies I

07/11/2017 Adaptive designs for DTA studies 8

[McCray et al., 2017; Alonzo et al., 2002]

Page 9: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

• Null hypotheses:  , : 1 , : 1

→ extendable to non‐inferiority

• Worst case scenario:

1 1 1 1

• Interim analysis based on  , , and π using maximum

likelihood estimation under a multinomial model

Sample size re‐estimation in pairedcomparative diagnostic accuracy studies II

07/11/2017 Adaptive designs for DTA studies 9

[McCray et al., 2017]

Page 10: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

Sample size re‐estimation in pairedcomparative diagnostic accuracy studies III

07/11/2017 Adaptive designs for DTA studies 10

[McCray et al., 2017]

Page 11: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

Sample size re‐estimation in pairedcomparative diagnostic accuracy studies IV

07/11/2017 Adaptive designs for DTA studies 11

[McCray et al., 2017]

Page 12: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

Application

• Comparison of CT and PET/CT for the diagnosis of pancreatic cancer

• Assumptions: π = 0.47

• Minimum sample size for TPPR = 0.71, TNNR = 0.46 → N = 186

• Interim analysis after 187 patients: TPPR = 0.79, TNNR = 0.64, π = 0.44

• Sample size re‐estimation: N = 275

07/11/2017 Adaptive designs for DTA studies 12

diagnostic test se sp

pre‐PET 81 % 66 %

post‐PET 90 % 80 %

Page 13: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

Further approaches

• Blinded sample size recalculation in clinical trials with binary 

composite endpoints (internal pilot study design) [Sander et al., 2017]

‐ two endpoints: composite endpoint and main component

‐ not co‐primary → at least one rejected null hypothesis required

• Blinded sample size re‐estimation in superiority and noninferiority

trials (internal pilot study design) [Friede et al., 2013]

‐ continuous outcome

‐ blinded estimation of the variance→ not possible for binary data

07/11/2017 Adaptive designs for DTA studies 13

Page 14: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

Structure

• Blinded interim analyses

• Unblinded interim analysis

• To look ahead

07/11/2017 Adaptive designs for DTA studies 14

Page 15: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

07/11/2017 Adaptive designs for DTA studies 15

[FDA, 2016]

Group sequential designs with / withoutsample size re‐assessment (e.g. based on corrected accuracy assumptions)

Adaptations regarding:‐ target population‐ comparator‐ hypotheses

Page 16: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

Approaches

• Interim evaluation of futility in clinical trials with co‐primary endpoints[Asakura et al., 2017]

‐ interim monitoring with predicted interval

‐ for normal distributed data → transferable to other scales

• Sequential designs for clinical trials with simultaneous bivariate response

(efficacy and safety) for normal distributed data[Todd, 2003; Jennison and Turnbull, 1993; Cook and Farewell, 1994]

07/11/2017 Adaptive designs for DTA studies 16

Page 17: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

Structure

• Blinded interim analyses

• Unblinded interim analysis

• To look ahead

07/11/2017 Adaptive designs for DTA studies 17

Page 18: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

07/11/2017 Adaptive designs for DTA studies 18

Adaptive study designs

Diagnosticaccuracy studies

Blindedinterim analysis

Adaptationsregarding:‐ prevalence‐ % discordant results‐ % missing values‐ reference standard‐ external: noninferiority margins,cutoff value

Unblindedinterim analysis

Adaptationsregarding:‐ estimated accuracy‐ target population‐ comparator‐ hypotheses

Randomizeddiagnostic studies

Blindedinterim analysis

Adaptationsregarding:diagnosticaccuracy orbenefit‐riskratio(external)

Unblindedinterim analysis

Adaptationsregarding:‐ proportion of TP, TN,FP, FN or benefit‐riskratio (internal)‐ study design

Adaptiveseamless designs

Page 19: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

Approaches

• Adaptive clinical trial designs for simultaneous testing of matched 

diagnostics and therapeutics [Scher et al., 2011]

• A conditional error function approach for subgroup selection in 

adaptive clinical trials [Friede et al., 2012]

• Biomarker adaptive designs in clinical trials [Chen et al., 2014]

• A Bayesian adaptive design for biomarker trials with linked 

treatments [Wason et al., 2015]

07/11/2017 Adaptive designs for DTA studies 19

Page 20: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

Further topics

• Adaptive seamless designs

• Development of R packages and SAS macros

• Next workshop in spring 2019?

• Webpage of the DFG‐project: http://www.ams.med.uni‐

goettingen.de/p‐mgmt/Flexh.html

07/11/2017 Adaptive designs for DTA studies 20

Page 21: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

Questions for discussion

07/11/2017 Adaptive designs for DTA studies 21

• Sample size reassessment based on interim estimates for the 

comparator? Blinded?

• Re‐estimation of the cutoff value?

→ training and validation dataset

• Modification of the reference standard?

• Modification of the target population – seamless design?

Page 22: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

References I• Alonzo et al. (2002). Sample size calculations for comparative studies of medical tests for 

detecting presence of disease. Stat Med, 21:835–52.• Asakura et al. (2017). Interim evaluation of futility in clinical trials with co‐primary endpoints. 

Talk on the CEN‐ISBS, Vienna. • Chen et al. (2014). Biomarker adaptive designs in clinical trials.  Transl Cancer Res, 3(3):279‐

292.• Cook et al. (1994). Guidelines for monitoring efficacy and toxicity responses in clinical trials. 

Biometrics 50:1146–1152.• FDA (2016). Adaptive designs for medical device clinical studies. Guidance for Industry and 

Food and Drug Administration Staff. https://www.fda.gov/downloads/ medicaldevices/deviceregulationandguidance/guidancedocuments/ucm446729.pdf (last access 5/11/17)

• Friede et al. (2013). Blinded sample size re‐estimation in superiority and noninferiority trials: bias versus variance in variance estimation. Pharmaceutical Statistics, 12(3):141–146.

• Friede et al.  (2012). A conditional error function approach for subgroup selection in adaptive clinical trials. Statistics in Medicine, 31(30):4309‐20.

• Jennison et al. (1993). Group sequential tests for bivariate response: interim analyses of clinical trials with both efficacy and safety endpoints. Biometrics 49:741–752.

22

Page 23: Adaptive designs for diagnostic accuracy studies - S5... · 07/11/2017 Adaptive designs for DTA studies 2 Although adaptive designs cannot ‘change the answer’ regarding the accuracy

References II

• Kairalla et al. (2012). Adaptive trial designs: a review of barriers and opportunities. Trials, 13:145.

• McCray et al. (2017). Sample size re‐estimation in paired comparative diagnostticaccuracy studies with a binary response. BMC Med Res Metthodol, 17:102.

• Rana et al. (2012). Clinical evaluation of an autofluorescence diagnostic device for oral cancer detection: a prospective randomized diagnostic study. Eur J Cancer Prev, 21(5):460‐466.

• Sander et al. (2017). Blinded sample size recalculation in clinical trials with binary composite endpoints. J Biopharm Stat. 2017;27(4):705‐715.

• Scher et al. (2011). Adaptive clinical trial designs for simultaneous testing of matched diagnostics and therapeutics. Clinical Cancer Research, 17(21):6634‐6640.

• Todd (2003). An adaptive approach to implementing bivariate group sequential clinicaltrial designs.

• Wason et al. (2015). A Bayesian adaptive design for biomarker trials with linked treatments. Br J Cancer, 113:699‐705.

23


Recommended