Models to Inform Recommendations About Preventive Services

1
Models to Inform Recommendations About
Preventive Services
Perspective of the United States Preventive
Services Task Force
2
An RCT for Every Topic is an Unattainable Goal
Technical

• Behavioral interventions
• inability to maintain fidelity with treatment
• inability to control provider effects
• Quality improvement interventions
• inconsistency of implementation
• difficulty in withholding interventions with what
  is viewed as high face validity
• contamination

3
An RCT for Every Topic is an Unattainable Goal
Technical

• Community programs
• Insufficient number of communities willing to be
  randomized
• Inability to maintain fidelity with program
  parameters
• Need for adaptability to gain cooperation
• Need for adaptability to achieve effectiveness
• Contamination

4
The Parachute Problem

• Do we really NEED an RCT?
• ....the effectiveness of parachutes has not
  been subjected to rigorous evaluation by using
  randomised controlled trials....we think that
  everyone might benefit if the most radical
  protagonists of evidence-based medicine organized
  and participated in a double-blind, randomized,
  placebo-controlled crossover trial of the
  parachute....
• Sackett. Parachute use to prevent death and
  major trauma related to gravitational challenge
  systematic review of randomised controlled
  trials. BMJ 20033271459-61.

5
But Even More Importantly

• It is not feasible to conduct an RCT for every
  topic for which clinical decisions must be made

6
CRC Options in 2008

• Something found in stool
• Occult blood using guaiac (gFOBT)
• Un-rehydrated
• Rehydrated
• Occult blood using test that is more sensitive
  than guaiac
• Fecal DNA
• Something seen in the colon
• Visual
• Flexible sigmoidoscopy
• Alone
• With FOBT (several options)
• Colonoscopy
• Radiographic
• Dual contrast barium enema (DCBE)
• CT colonography (CTC)

7
Number of RCTs CRC Screening Type of Test

• 9 strategies against no screening
• Would require 8 two-arm trials or a very large
  8-arm trial (unrehydrated gFOBT already done)
• 9 strategies against all other possible
  strategies alone
• E.g. more sensitive FOBT versus unrehydrated FOBT
• FOBT versus optical colonoscopy
• optical colonoscopy versus CT colonography
• flexible sigmoidoscopy versus fecal DNA
  etc. etc. etc.
• Would require 35 two-arm trials or a very large
  8-arm trial

8
It May be Impossible or Nearly Impossible To
Conduct a Valid RCT

• Starting age
• Age 40 or 50 or something else for CRC
• Age 16 or 21 or something else for cervical
  cancer
• Stopping age
• Stop age 75 versus 85 for CRC
• Stop age 65 versus 75 for cervical cancer
• Screening interval
• Every 5 years versus every 10 years versus once
  in a lifetime for CRC
• Every year versus every 3 years versus every 5
  years for cervical cancer

9
Models What If.....

• Long-Term Financial Planning
• What if I live to age 95, how much money will I
  need to live comfortably considering inflation
  and the rate of return on my investments? (MAKE A
  MODEL)
• What do you recommend to assure that I dont run
  out of money? (USE THE MODEL TO INFORM THE
  RECOMMENDATION)

10
Models What If.....

• Short-Term Financial Decisions
• What if I buy a car that gets 65 miles per
  gallon, how much money will I save over five
  years considering the number of miles I drive?
  (MAKE A MODEL).
• Should I buy a car that costs 34,000? (USE THE
  MODEL TO INFORM THE DECISION)

11
Models What If.....

• Policies
• What if people in developing economies use the
  same number of barrels of crude oil per capita as
  people in the United States, in what year will
  the world supply of crude oil be exhausted?
  (MAKE A MODEL)
• What does the committee recommend to assure that
  there is sufficient availability of energy for
  future generations? (USE THE MODEL TO INFORM THE
  RECOMMENDATION)

12
Models What If.....

• Daily Decisions
• What if I leave the office at 3 oclock instead
  of 5 oclock, how long will it take to get back
  home over the Bay Bridge considering the month,
  the day of the week and the weather? (MAKE A
  MODEL).
• Should I leave early today to be sure that I am
  not late to dinner at the Smiths? (USE THE MODEL
  TO INFORM THE DECISION)

13
Models in the Context of Evidence-Based Medicine

• Conscientious careful, thorough, meticulous
• Explicit open, clear, plain
• Judicious sensible, well thought out

Good models meet this criterion
Modelers need to do this better
Good models meet this criterion
14
Results From Good Models Does This Information
Help Make a Recommendation? The USPSTF Thinks
the Answer is Yes.
15
Results From Good Models Does This Information
Help Make a Recommendation? The USPSTF Thinks
the Answer is Yes.
16
Models the Pushback

• Not taught in most medical schools
• Generalized antipathy many people hate anything
  called a model
• Assumptions are made these often look like
  guesses
• The model assumptions are often opaque

17
MODELS

• Information from models reduces bounds estimates
• Information from models is better than
  information obtained by guessing, even informed
  guessing
• A recommendation based on a good model is better
  than silence and is often the best we will ever do

18

• All models are wrong, some are useful.
• George Box
  1979

19

• DISCUSSION