Modular Programs Sebastian Schneeweiss, MD, ScD Jennifer Nelson, - - PowerPoint PPT Presentation

▶
modular programs
SMART_READER_LITE
LIVE PREVIEW

Modular Programs Sebastian Schneeweiss, MD, ScD Jennifer Nelson, - - PowerPoint PPT Presentation

Aug 31, 2023 390 likes •580 views

Modular Programs Sebastian Schneeweiss, MD, ScD Jennifer Nelson, PhD Mini-Sentinel Methods Core January 31, 2013 info@mini-sentinel.org 1 Modular approach to drug safety monitoring in a distributed database system Principal idea

slide-1
SLIDE 1

info@mini-sentinel.org 1

Modular Programs

Sebastian Schneeweiss, MD, ScD Jennifer Nelson, PhD Mini-Sentinel Methods Core January 31, 2013

slide-2
SLIDE 2

info@mini-sentinel.org 2

Modular approach to drug safety monitoring in a distributed database system

 Principal idea

  • Pre-programmed modules can be quickly activated to run adjusted

analyses across data partners

  • For monitoring, modules will be run repeatedly as data are refreshed

 Some specifications

  • Validated programming code
  • Can be run asynchronously across data partners as data get refreshed

while preserving data privacy

  • Confounding adjustments via self-controlled designs, PS matching or

regression analyses

  • Estimate ratio and difference measures (rate or risk)
  • Sequential (or group sequential) analyses
slide-3
SLIDE 3

info@mini-sentinel.org 3

Define exposures, outcomes, etc Estimate the risk Newly marketed product Choose analysis approach

Prospective surveillance: estimate risk

Module 1

Self-controlled

Module 2

Cohort matching

Module 3

Cohort regression

Risk estimation

slide-4
SLIDE 4

info@mini-sentinel.org 4

Self-controlled Key parameters:

  • Exposure crossover
  • Risk and control

window

  • Exposure time trend

adjustment

Cohort Matching Key parameters:

  • Score-based

matching (PS, DRS)

  • hd-PS
  • 1:1/variable ratio
  • AT vs. ITT

Cohort Regression Key parameters:

  • Regression
  • IPT weighted

regression

  • Tailored to the rare

event setting

Module 1

Module 2

Module 3 Cohort identification (MP3) Index identification

Prospective surveillance: estimating risk

Sensitivity analyses

  • Popn. Subgroups
  • Dose subgroups
slide-5
SLIDE 5

info@mini-sentinel.org 5

Coordinating center

  • Identify Cohort,
  • Outcomes
  • Covariates
  • Calculate

confounder scores (PS, hd-PS, DRS) Specify input parameters

  • Run diagnostics
  • Create de-identified

result files Evaluate diagnostics and aggregate data across partners Apply alerting algorithms and interpret results Iterate at next data refresh

Multiple data partners

Transmit code Transmit data

Start Module 2

Module 2 in detail

slide-6
SLIDE 6

info@mini-sentinel.org 6

Diagnostics: Balance before matching

Table 1. Cohort of New Initiators of Rofecoxib and Non-Selective NSAID (Unmatched) N (%) N (%) Absolute Difference Standardized Difference Characteristic rofecoxib nsaid Number of patients 9409 (100.0 %) 9977 (100.0 %) Number of Events While on Therapy 39 (0.4 %) 15 (0.2 %) Person time at risk 59.9 ( 33.3) 46.4 ( 32.5) Patient Characteristics Age 76.3 ( 10.7) 73.1 ( 12.2) 3.2 3.2 60-70 1305 (13.9 %) 1679 (16.8 %)

  • 2.9
  • 0.082

70-80 3631 (38.6 %) 3883 (38.9 %)

  • 0.3
  • 0.007

80-90 3179 (33.8 %) 2619 (26.3 %) 7.5 0.164 90-100 580 (6.2 %) 395 (4.0 %) 2.2 0.101 Gender (F) 7764 (82.5 %) 7374 (73.9 %) 8.6 0.208 Recorded use of: Ace Inhibitors 1224 (13.0 %) 1351 (13.5 %)

  • 0.5
  • 0.016

ARB 567 (6.0 %) 535 (5.4 %) 0.6 0.029 Anticoagulants 548 (5.8 %) 328 (3.3 %) 2.5 0.122 Primary Analysis Covariate Balance And many more …

… … …

slide-7
SLIDE 7

info@mini-sentinel.org 7

Diagnostics: Balance before/after matching

slide-8
SLIDE 8

info@mini-sentinel.org 8 10 5 90 95 100 100 15 10 185 190 200 200

DP1 DPn Launch date + 3 mos.

25 15 275 285 300 300

.

. .

Data aggregation across data partner

E E

_

D D

_

slide-9
SLIDE 9

info@mini-sentinel.org 9 10 5 90 95 100 100 15 10 185 190 200 200

DP1 DPn Launch date + 3 mos. + 3 mos.

25 15 275 285 300 300

+ 3 mos.

15 10 185 190 200 200 30 20 370 380 400 400 45 30 555 570 600 600 25 20 275 280 300 300 40 30 560 570 600 600 65 50 835 850 900 900

+ 3 mos.

50 40 450 460 500 500 100 75 900 925

1000 1000

150 115

1350 1385 1500 1500

+ 3 mos. .

. .

Data aggregation across DPs & over time

E E

_

D D

_

slide-10
SLIDE 10

info@mini-sentinel.org 10 10 5 90 95 100 100 15 10 185 190 200 200

Launch date + 3 mos. + 3 mos.

25 15 275 285 300 300

+ 3 mos.

100 75

1000 1025 1100 1100

185 135

2015 2065 2200 2200

285 210

3015 3090 3300 3300

30 20 370 380 400 400 30 20 370 380 400 400 40 30 460 470 500 500 40 30 560 570 600 600 80 60 1020 1040 1100 1100

+ 3 mos.

50 40 450 460 500 500 100 75 900 925

1000 1000

150 115

1350 1385 1500 1500

+ 3 mos. DP1 DPn .

. .

Asynchronous database refreshes

E E

_

D D

_

10

slide-11
SLIDE 11

info@mini-sentinel.org 11 10 5 90 95 100 100 15 10 185 190 200 200

Launch date + 3 mos. + 3 mos. + 3 mos.

15 10 185 190 200 200 30 20 370 380 400 400 25 20 275 280 300 300 40 30 560 570 600 600

+ 3 mos.

50 40 450 460 500 500 100 75 900 925

1000 1000

+ 3 mos.

Visualizing heterogeneity

E E

_

D D

_ Heterogeneity by time since marketing Center effects DP1 DPn .

. .

11

slide-12
SLIDE 12

info@mini-sentinel.org 12

Data aggregation (Report: Rassen et al.)

Rassen JA, et al. Pharmacoepidemiol Drug Saf 2010;19:848-57

  • 10.0
  • 8.0
  • 6.0
  • 4.0
  • 2.0
0.0 2.0 4.0 6.0 8.0 10.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
slide-13
SLIDE 13

info@mini-sentinel.org 13

Aggregation over time

… … … … … … …

PS-match PS-match PS-match

slide-14
SLIDE 14

info@mini-sentinel.org 14

Risk estimation Aggregate accumulating results over time Apply alerting rules

Module 1 Self-controlled Module 2 Cohort matching Module 3 Cohort regression Sensitivity analyses

Prospective surveillance: alerting

slide-15
SLIDE 15

info@mini-sentinel.org 15

Pre-monitoring activities

 Acceptable false positive rate may vary:  Acceptable level of risk

Availability of alt. meds Severity of event(s) Expected beneficial effect

 Anticipated utilization

Monitoring intervals Duration of monitoring

slide-16
SLIDE 16

info@mini-sentinel.org 16

Post-monitoring activities

 Sensitivity analyses

Confounding Exposure risk-window Incident user definition window AT vs. ITT

 Subgroup analyses as needed  Comprehensive presentation of decision-relevant

information

slide-17
SLIDE 17

info@mini-sentinel.org 17

Define exposures, outcomes, etc Estimate the risk Aggregate results over time Apply alerting rules Report to FDA FDA reports to public when appropriate Newly marketed product Choose analysis approach

Prospective surveillance: reporting

slide-18
SLIDE 18

info@mini-sentinel.org 18

What happens when we find something?

 Examples of follow-up

activities:

  • Data validity checks, analytic code

checks

  • Adjust for additional confounders
  • Test against other comparators
  • Medical chart validation of cases
  • Quantitative bias analysis
  • Detailed epidemiologic investigation

to assess causality

 Prompt, pre-planned product-specific assessment of

positive signal