James Burgess, PhD Center for Healthcare Organization and - - PowerPoint PPT Presentation

Martin Kulldorff, PhD

Division of Pharmacoepidemiology and Pharmacoeconomics, Brigham and Women’s Hospital, Harvard Medical School

Robert Penfold, PhD

Group Health Research Institute and Department of Health Services Research, School of Public Health, University of Washington

James Burgess, PhD

Center for Healthcare Organization and Implementation Research and Professor, Dept. of Health Law, Policy and Management, Boston University School of Public Health

Sharon Greene, PhD, MPH

Bureau of Communicable Disease, New York City Department of Health and Mental Hygiene and Mailman School of Public Health, Columbia University

Introduction

 Brief overview of methods

 Why use space-time cluster analysis?  Calculation of a scan statistic

 Three case studies

 Retrospective analysis of new psychotropic medication use

among Medicaid youth in Michigan

 Retrospective analysis of the spread of new antipsychotic

prescribing for Bipolar Disorder in the VA

 Prospective cluster detection of GI illness in Northern

California and of reportable diseases in New York City.

 Discussion/Q&A about use of space-time cluster analysis

Why Use a Scan Statistic?

With “outbreaks”:

 We do not know where they will occur.  We do not know their geographical size.  We do not know when they will occur.  We do not know how rapidly they will

emerge/spread.

Importance of Early Outbreak Detection

 Earlier identification of new cases  Quarantine cases  Warn about risk factors  Eliminate health hazards and low value care  Scientific research concerning policies and treatments

Space-Time Scan Statistic

Use a cylindrical window, with the circular base representing space and the height representing time.

1. For each cylinder:

• Obtain actual and expected number of cases inside

and outside the cylinder.

• Calculate likelihood function.
• 2. Compare Cylinders:
• Pick cylinder with highest likelihood function as

most likely cluster.

• 3. Inference:
• Generate random replicas of the data set under

the null-hypothesis of no clusters (Monte Carlo sampling).

• Compare most likely clusters in real and random

data sets (Likelihood ratio test).

A Space-Time Permutation Scan Statistic for Disease Outbreak Detection

Kulldorff M, Heffernan R, Hartman J, Assunção R, Mostashari F. A space-time permutation scan statistic for disease outbreak detection. PLoS Medicine, 2005, 2:216-224.

• Cases only
• To be used when there is no reliable denominator

data or;

• When rate-based inference is not appropriate

Space-Time Permutation Scan Statistic

• 1. For each cylinder, calculate the expected

number of cases conditioning on the marginals where czd = # cases in location z at time d and C = total number of cases

 Expected number of cases in cylinder A is the sum of

all space-time days within that cylinder

2.

The probability of a particular observed count

• ccurring is given by the hypergeometric

distribution

Space-Time Permutation Scan Statistic

3.

When the sum of cases for each spatial and temporal unit is small compared to the total number of cases, the Poisson distribution can be used to approximate (4) and we can use the Poisson Generalized Likelihood Ratio to measure evidence of an

• utbreak:

Space-Time Permutation Scan Statistic

 Of all cylinders evaluated, the one with the greatest

GLR is the least likely to occur by chance.

Space-Time Permutation Scan Statistic

• 4. Generate random replicas of the data set

conditioned on the marginals, by permuting the pairs of spatial locations and times.

• 5. Compare test statistic in real and random

(simulated) data sets using Monte Carlo hypothesis testing (Dwass, 1957): p-value = R(GLRmax) / (1+ # simulations)

Space-Time Permutation Scan Statistic

Space-Time Permutation Scan Statistic: Properties

• Adjusts for purely geographical clusters.
• Adjusts for purely temporal clusters.
• Simultaneously tests for outbreaks of any size at

any location, by using a cylindrical windows with variable radius and height.

• Accounts for multiple testing.
• Aggregated or non-aggregated data (counties, zip-

code areas, census tracts, individuals, etc).

Other Probability Models

 Counts:

 Bernoulli: cases vs. controls  Discrete Poisson  Space-time permutation

 Multinomial, Ordinal  Exponential: survival time  Normal: continuous distribution  Satscan Users Guide



http://www.satscan.org/cgi-bin/satscan/register.pl/SaTScan_Users_Guide.pdf?todo=process_userguide_download

Health Care Policy Questions

 Were individuals in some places more likely to have

___ occur after implementation of policy _____?

 Are individuals in some places doing ____ at a faster

rate than other places?

 If we conduct real-time surveillance, could we target

resources to stop ____ from happening?

What do I need?

 A good hypothesis and plausible contagion

mechanism 

 Event data: new infections, new diagnoses, new

prescriptions, new injuries, . . .

 Time and location for every event

 Unit of time or space does not matter

 Download SatScan (free) http://www.satscan.org/