1. Advanced Disease Identification Electronic Medical Record Data
with
Electronic Medical Record Data
Public Health Information Network Conference
August 28, 2008
Michael Klompas MD, MPH, FRCPC
CDC Center of Excellence in Public Health Informatics
Harvard Medical School and Harvard Pilgrim Health Care, Boston, MA

2. CDC Center of Excellence in Public Health Informatics (Boston)
Harvard Medical School / Harvard Pilgrim Health Care Department of Ambulatory Care and Prevention
Children’s Hospital Informatics Program
Massachusetts Department of Public Health
Harvard Vanguard Medical Associates (for Atrius Health)
Brigham and Women’s Hospital Channing Laboratory 2

3. Overview
Overview of Electronic medical record Support for Public Health (ESP)
Available data elements in ESP
Sample disease identification algorithms
Algorithm development example: acute hepatitis B

4. Electronic Support for Public health (ESP)
Software and architecture to automate detection and reporting of notifiable diseases
Surveys codified electronic medical record data for patients with notifiable conditions
Generates and sends secure electronic case reports to the state health department
MMWR 2008;57: Advances Disease Surveillance 2007;3:3 4

5. ESP: Automated detection and reporting of notifiable conditions
Practice EMR’s
ESP Server
D P H
Health Department
diagnoses
lab results
meds
demographics
vital signs
HL7 electronic case reports of notifiable conditions
I added “vital signs”

6. ESP Operational in Atrius Health January 2007 to present
~27 multispecialty ambulatory practices in MA
EpicCare EMR
~600,000 patients
750 clinicians
An ESP server resides in the central data processing center
Analyzes data from all 27 sites
Boston, MA
I added “vital signs”
© Google Maps 6

7. Available Data Elements in ESP
Category
Format
Demographic Data
Carat-delimited text
Vital signs
Encounter types / dates
Diagnostic codes
ICD9
Lab test orders
CPT codes mapped to LOINC
Lab test results
Prescriptions
NDC codes and generic names
Organism names
SNOMED

8. Electronic Support for Public Health (ESP)
Currently reports:
Chlamydia
Gonorrhea
Pelvic inflammatory disease
Acute hepatitis A
Acute hepatitis B
Acute hepatitis C
Active tuberculosis
Under validation:
Syphilis
Measles
Mumps
Chronic hepatitis B

9. Algorithm Accuracy Atrius Health, June 2006-present
Condition
Total Cases
False Positives
Positive Predictive Value
Chlamydia
1790
100%
Gonorrhea
224
Pelvic inflammatory disease 39 1
97%
Acute hepatitis A 6
83%
Acute hepatitis B 15
Acute hepatitis C 14
Tuberculosis 5
67%

10. ESP vital signs orders ICD9’s lab results meds chlamydia
active tuberculosis
acute hepatitis B

11. Case Identification Logic: Chlamydia
Positive test for any of the following:
Test Name
CPT
Component
LOINC
CHLAMYDIA PCR, URINE (MALES
86631
835
CHLAMYDIA TRACHOMATIS CULTURE
87110
3474
6349-5
CHLAMYDIA GENPROBE DNA
87491
1312
PEDIATRIC URINE CHLAMYDIA
2487
CHLAMYDIA TRACHOMATIS DNA, SDA
2801
CHLAMYDIA TR DNA
2878
CHLAMYDIA TR DNA URN
2879
CHLAMYDIA TR. DNA
2906
CHLAMYDIA TRACHOMATIS, DNA PROBE, FEMALE
4312
CHLAMYDIA TRACHOMATIS, DNA, SDA
4320
CHLAMYDIA TRACHOMATIS
4803
87591
URINE GC AND CHLAMYDIA, PEDIATRIC BY APT
2686
CHLAMYDIA & GC WITH REFLEX TO IDENTIFICATION
87800
4310

13. What about diseases that cannot be diagnosed from a lab result alone?
examples: culture-negative tuberculosis acute hepatitis B acute hepatitis C pelvic inflammatory disease early lyme disease early syphilis

14. Lab diagnosis (positive culture or PCR)
Limitations of diagnosis by labs or ICD9s alone the example of tuberculosis
Lab diagnosis (positive culture or PCR)
Excellent positive predictive value
But misses culture negative cases (~20% of all TB cases)
ICD9 based surveillance
Similar sensitivity to lab-based surveillance
Also misses ~20% of cases
Poor positive predictive value (~15%)
False positives: remote history of tuberculosis or +PPD

15. Solution
Integrate multiple streams of data from the EMR to increase sensitivity and specificity
Lab orders
Lab results (present and past)
Medication prescriptions
ICD9 diagnoses

16. Case Identification Logic Active Tuberculosis
Any of the following:
Prescription for pyrazinamide OR
Order for (AFB smear or culture) and an ICD9 code for TB within 60 days
Order for 2 or more anti-tuberculous medications and an ICD9 code for TB within 60 days

17. Tuberculosis algorithm validation
Algorithm applied to Atrius Health, June 2006 to present
Identified 15 cases
10/15 ultimately confirmed to have active disease
9 of the 10 previously reported to health department
1 case previously unknown to health department
No cases known to health department missed by ESP
False positives - suspected cases of active TB that don’t pan out
Woman with 3 months cough after travel to India, started on empiric TB therapy, ultimately diagnosed with Mycobacteria gordonae

19. Developing Case Identification Algorithms
The example of Acute Hepatitis B

20. Is this acute or chronic hepatitis B?
The Problem
How does one electronically distinguish acute hepatitis B from flares of chronic hepatitis B?
Scenario:
33 year old gent from Vietnam presenting with fatigue
Liver enzymes elevated to the mid-600’s
Viral hepatitis panel positive for hep B surface antigen
Is this acute or chronic hepatitis B?

21. CDC Definition for Acute Hepatitis B
An acute illness with discrete onset of symptoms and
Jaundice or elevated serum aminotransferase levels and
IgM antibody to hepatitis B core antigen positive OR hepatitis B surface antigen (HBsAg) positive

22. Acute hepatitis B Development of a detection algorithm
Derivation:
Begin with CDC case definition
Translate into electronic terms
Apply to historic data, Harvard Vanguard,
Review charts of all identified patients
Refine algorithm based on chart reviews
Validation:
Apply to prospective data, Atrius Health,
Compare findings with health department records
Total number of cases identified
Missed cases
repeat

23. Acute hepatitis B Development of a detection algorithm
Strategy 1: ICD9
Random selection of 50 patients with ICD within the past two years
Charts reviewed
Positive Predictive Value
0% (95% confidence interval, 0-6%)

24. Acute hepatitis B Development of a detection algorithm
Strategy 2: current lab tests
ALT or AST > 5x normal AND
Positive hepatitis B surface antigen
Positive Predictive Value
47% (95% confidence interval, 41-53%)

25. Acute hepatitis B Development of a detection algorithm
Strategy 3: current & past lab tests & ICD9’s
ALT or AST > 5x normal AND
Positive hepatitis B surface antigen AND
No prior positive hepatitis B surface AND
No ICD9 code for chronic hepatitis B ever
Positive Predictive Value
68% (95% confidence interval, 61-75%)

26. Acute hepatitis B Development of a detection algorithm
Strategy 4: current & past lab tests & ICD9’s
ALT or AST > 5x normal AND
Positive hepatitis B surface antigen AND
No prior positive hepatitis B surface AND
No ICD9 code for chronic hepatitis B ever AND
Total bilirubin >1.5
Positive Predictive Value
97% (95% confidence interval, %)
Sensitivity 99% Specificity 94%

27. Acute Hepatitis B Algorithm Validation
Final algorithm applied to ESP’s prospective dataset
June 2006 to April 2008
8 cases identified
Chart review:
100% positive predictive value
Comparison to MA Dept of Health Records
Only 1 acute case known to health department
3 of the known cases misclassified by DPH as chronic
4 previously unreported cases found by ESP
No missed cases
PLoS ONE 2008;3:e2626

29. Sorting through positive Hep B Results - ESP versus ELR
601 distinct patients
8 acute
593 chronic cases
2648 positive test results for hepatitis B
E S P
E L R

30. EMR algorithms for disease detection
Surveillance using lab data alone misses cases and misclassifies others
Algorithms combining multiple current and prior lab tests, ICD9 codes, and med orders can identify complex conditions with high accuracy
Validation labour intensive but essential
Standardization of electronic case definitions could make surveillance more efficient, complete, and comparable across states
Community of practice opportunity?

31. Contact: mklompas@partners.org
ESP Team
Harvard Medical School / Harvard Pilgrim Health Care Department of Ambulatory Care and Prevention
Richard Platt MD, MSc  Ross Lazarus MBBS, MPH, MMed
Julie Dunn MPH  Michael Calderwood MD
Ken Kleinman ScD  Yury Vilk PhD
Kimberly Lane MPH
Harvard Vanguard Medical Associates
Francis X. Campion MD
Benjamin Kruskal MD, PhD
Massachusetts Department of Public Health
Alfred DeMaria MD  Bill Dumas RN
Gillian Haney MPH  Daniel Church MPH
James Daniel MPH  Dawn Heisey MPH
Channing Laboratory of Brigham and Women’s Hospital
Xuanlin Hou MSc
Collaborators Wanted!
Contact: