Evaluation of Electronic Databases for Reportable Disease Case Identification

1
Evaluation of Electronic Databases for Reportable
Disease Case Identification

• Gosia Kubiak, MSc, MPH
• Wendi Bowman, MPH
• Christopher Rennix, ScD, CIH
• Navy Environmental Health Center, EpiData Center
• Syndromic Surveillance Conference October 11,
  2007

2
Purpose

• To improve medical surveillance through
  utilization of existing electronic records in the
  Military Health System (MHS)
• Evaluate the ability to establish an automated
  surveillance system based on a combination of
  electronic medical databases
• Laboratory Health Level7 (HL7)
• Ambulatory care Standard Ambulatory Data Record
  (SADR)
• Inpatient care Standard Inpatient Data Record
  (SIDR)
• Naval Disease Reporting System (NDRS)
• Passive disease reporting

3
Methodology

• Study was based on a pilot sample of data
• 01 Jan 2006 31 Dec 2006
• Camp Lejeune, North Carolina
• Active duty members
• Navy, Marine Corps
• Positive Chlamydia trachomatis
• Case definition
• Infection with C. trachomatis
• ICD-9 099.41, 078.8
• Isolation by culture or detection of an antigen

4
Methodology cont.

• All analysis performed in SAS (v. 9.1)
• Established program in SAS to check for data
  validity, case definition, and case
  deduplification
• Matching
• Match cases by the unique identifier, potential
  MM match
• Capture-Recapture Method
• Estimation of the total number of cases that
  were not identified by all three databases
• Generalized Linear Models
• Interaction effects between the databases
• Hierarchical structure

5
Capture Recapture Method

• Each case should have a chance to be listed on
  each of the data sources
• Difficult for medical data sources
• e.g. Inpatient data Death Records (severity
  dependent)
• e.g. Physician referrals Admissions (physician
  dependent)
• Mutually exclusive data sources not applicable to
  method
• Entirely independent no overlap
• Best to use 3-5 data sources
• Dependence can be evaluated
• Method more robust for multiple sources
• Matching
• Exact matching
• Probabilistic matching
• Combination matching

6
Capture Recapture Method (2)

• If using K sources
• 2k data points
• 2K possible GENMOD models
• K sample model assumption
• No K-way interaction or higher
• Disadvantage to using a log linear model for K
  data sets
• Cannot attribute characteristics of differences
  in capture
• Covariates can be included
• Goodness of Fit testing
• Confidence Interval for estimate

7
Capture Recapture Method (3)

• Strengths
• Conservative estimate
• Use of existing data sources cost effective,
  timely vs. case counting
• Relatively robust
• Supplement to knowledge of incidence rates
• Limitations
• Repeat capture not always available
• Reasons for why persons are not captured
• Not seeking care
• Not as assessment of disease prevalence only
  identified disease
• Does not account of false positive results

8
Results
76
HL7/SADR/Both, in NDRS 80 60
64
No Chlamydia cases were identified in SIDR
9
67
33
HL7 only SADR only (HL7 SADR)
10
Results (2)

• There were no differences between the types of
  cases identified from each source
• Description of cases
• 93 Marine Corps v. 7 Navy
• 98 Enlisted
• 67 Male
• 66 Caucasian
• 67 SADR cases Chlamydia-related primary dx

11
Results (3)

• Process specific issues were identified which may
  explain the discrepancy of reported cases and no
  electronic record.
• Operational clinics may report directly to
  Preventative Medicine Dept but do not have access
  to CHCS system
• Strong PM presence at the site may have resulted
  in increased reporting.
• Electronic case identification for Camp Lejeune
  MTFs
• 74 of Chlamydia cases in 2006 would be
  captured
• Operational Clinics
• Alignment of the encounter/lab data with NDRS
  depends on command impact
• Electronic case identification may be efficient
  for Defense Health Provider facilities, but not
  operational clinics.

12
Project/Result Limitations

• Results were limited to data in electronic
  clinical records
• Results for Camp Lejeune treatment facilities
  only
• Choice of Chlamydia was based on disease
  prevalence
• A lot of cases, therefore easier to capture
• Method may provide estimates with greater
  variance for rare or infrequent diseases
• Method may prove effective in removing the burden
  of disease reporting by providers/commands for
  population-wide or prevalent diseases
• E.g. Chlamydia, Gonorrhea, Hepatitis
• Project identified CY2005 as a system-transition
  year

13
Conclusions

• Based on pilot site, 3 out of 4 Chlamydia cases
  were captured through electronic databases
• Case burden versus population burden
• Ability to estimate denominator
• gt Calculate disease burden, command specific
• Method may aid in the removal of reporting burden
  for prevalent diseases,
• Allow to address rare or mission-impacting cases
• Majority of cases identified
• Captured cases are representative of the
  population

14
QUESTIONS?

• Thank you

Contact Gosia Kubiak 757.953.0979 Gosia.Kubiak_at_me
d.navy.mil