The Space-Time Check Measurement for Different Information Streams

Presentation Transcript

The Space-Time Scan Statistic for Multiple Data Streams Martin Kulldorff, Katherine Yih, Ken Kleinman, Richard Platt, Harvard Medical School and Harvard Pilgrim Health Care Farzad Mostashari, New York City Department of Health and Mental Hygiene Luiz Duczmal, Univ Fed Minas Gerais, Brazil

Different Data Sources For instance: OTC Drug Sales, from drug store chains Nurses Hotline Calls, from Optum Regular Physician Visits, from HMOs/VA Emergency Department Visits, from doctor's facilities Ambulance Dispatches, from 911 call focuses Lab Test Results, from labs

Different Types of Data from the Same Data Source For instance, HMO information concerning: Telephone Calls to Physicians Regular Physician Visits Emergency Department Visits Lab Test Requests Lab Test Results Drug Prescriptions

Different Groupings in the Same Type of Data Children, Young Adults, Adults age 65+ Male, Female Diarrhea, Vomiting

Early Work Burkom HS, Biosurveillance Applying Scan Statistics with Multiple, Disparates Data Sources, Journal of Urban Health, 80i:57-65, 2003 Wong WK, Moore A, Cooper G, Wagner M. WSARE: What's unusual about late occasions? Diary of Urban Health, 80i:66-75, 2003.

Why Multivariate Detection Methods? We don't know whether an episode will make a flag in at least one information streams. The enlightening substance is diverse in various information streams.

Outline Method: Space-Time Permutation Scan Statistic Example: Gastrointestinal phone calls, pressing consideration visits and normal doctor visits in Boston

The Spatial Scan Statistic Create a general or unpredictable matrix of centroids covering the entire study area. Make a limitless number of circles around every centroid, with the range anywhere in the range of zero up to a greatest so that at most 50 percent of the populace is incorporated.

A little example of the circles utilized

Space-Time Scan Statistic Use a round and hollow window, with the roundabout base speaking to space and the tallness speaking to time. We will just consider chambers that achieve the present time.

Space-Time Permutation Scan Statistic 1. For every barrel, figure the normal number of cases molding on the marginals μ st = C s C t/C where C s = # cases in area s C t = # cases in time interim t C = add up to number of cases

Space-Time Permutation Scan Statistic Let c st = # cases in the chamber covering area s and time interim t.

Space-Time Permutation Scan Statistic 2. For every chamber, ascertain the Poisson probability T st = [c st/μ st ] c st x [(C-c st )/(C-μ st )] C-c st if c st/μ st > 1, T st = 1 generally 3. Test measurement T = max st log [ T st ]

Statistical Inference 4. Create arbitrary imitations of the information set molded on the marginals, by permuting the sets of spatial areas and times. 5. Think about test measurement in genuine and arbitrary information sets utilizing Monte Carlo speculation testing (Dwass, 1957): p = rank(T genuine )/(1+#replicas)

Multiple Data Streams For every chamber, include the Poisson log probabilities: T st = log[ T [1] st ] +log[ T [2] st ] +log[ T [3] st ] Test measurement T = max st T st

Syndromic Surveillance in Boston: Upper and Lower GI Harvard Pilgrim Health Care HMO individuals watched over by Harvard Vanguard Medical Associates Historical Data from Jan 1 to Dec 31, 2002 Mimicking Surveillance from Sept 1 to Dec 31, 2002

Three Data Streams Telephone Calls ( ~ 20/day) Urgent Care Visits ( ~ 9/day) Regular Physician Visits ( ~ 22/day) Multiple contacts by a similar individual evacuated.

Strongest Signal: October 18 p= Recurrence Int. Tele: 0.001 < 1/1000 days Urgent 0.91 ~ consistently Regular: 0.84 ~ consistently Multiple DS: 0.001 < 1/1000 days

October 18 Signal Friday Number of Cases: 5 Expected Cases: 0.04 Location: Zip Code 01740 Time Length: One Day

October 18 Signal Friday Number of Cases: 5 Expected Cases: 0.04 Location: Zip Code 01740 Time Length: One Day Diagnosis: Pinworm Infestation (every one of the 5)

October 18 Signal Friday Number of Cases: 5 (all tele) Expected Cases: 0.04 Location: Zip Code 01740 Time Length: One Day Diagnosis: Pinworm Infestation (each of the 5) Same Family: Mother, Father, 3 Kids

2 nd Strongest Signal: December 20 p= Recurrence Int. Tele: 0.03 1/32 days Urgent 0.71 ~ consistently Regular: 0.003 1/333 days Multiple DS: 0.002 1/500 days

December 20 Signal Number of Cases: 16 (7 tele, 7 normal, 2 pressing) Expected Cases: 3.5 Location: Zips 01810,26,45,50,52,76 Time Length: Two Days (Thu, Fri) Strong flags on the two after days.

December 20 Signal Mostly various ambiguous GI analyze: Esophageal Reflux (3), Nausea (2), Abdominal Pain (2), Noninfectious GI (2), Acute pharyngitis, Mastodynia, Diarrhea, Anemia, Hypertension, Blood in stool, Holiday parties?

3 rd Strongest Signal: October 26 p= Recurrence Int. Tele: 0.07 1/14 days Urgent 0.85 ~ consistently Regular: 0.18 1/6 days Combined: 0.007 1/142 days

October 26 Signal Saturday Number of Cases: 8 ( 5 tele, 3 customary) Expected Cases: 0.9 Location: Zip Codes 01902,07,15,45,70 Time Length: Two Days (Fri, Sat) Various particular findings.

Research Funded By Methods: Alfred P Sloan Foundation Data, National Bioterrorism Syndromic Surveillance Demonstration Program: National Center for Infectious Diseases, Centers for Disease Control and Prevention

Free Software SaTScan v 5.1 www.satscan.org