mam ibraheem, md, mph
DESCRIPTION
Using Hospitalization Data To Evaluate and Improve Invasive Pneumococcal Disease Surveillance — New Mexico, 2007–2009. Mam Ibraheem, MD, MPH. New Mexico Department of Health EIS Field Assignments Branch, DAS, SEPDPO, OSELS 2011 CSTE Annual Conference June 15, 2011. - PowerPoint PPT PresentationTRANSCRIPT
Mam Ibraheem, MD, MPHNew Mexico Department of Health
EIS Field Assignments Branch, DAS, SEPDPO, OSELS2011 CSTE Annual Conference
June 15, 2011
Using Hospitalization Data To Evaluate and Improve Invasive Pneumococcal Disease Surveillance —
New Mexico, 2007–2009
Office of Surveillance, Epidemiology, and Laboratory ServicesScientific Education and Professional Development Program Office
Invasive Pneumococcal Disease (IPD)
Isolation of Streptococcus pneumoniae from normally sterile sites
Serious and vaccine-preventable Typically manifests as pneumonia, septicemia, or meningitis Leading cause of bacterial meningitis in young children in
the United States
Importance of IPD Surveillance Systems
Monitor pneumococcal vaccination programs Monitor changes in IPD epidemiology Inconsistent reporting adversely impacts policy decisions
IPD Surveillance in New Mexico
IPD reportable since 2000 Passive surveillance
Statewide Healthcare providers/Laboratories
Active Bacterial Core Surveillance (ABCs) Population-based: cases among non-residents of NM excluded Audits of clinical laboratory records used to identify cases not
reported passively In 2009, access to hospitalization data
Questions
How complete is the combined (passive and active) IPD surveillance in New Mexico?
Can hospitalization data identify additional IPD cases?
Capture-Recapture Method
Degree of undercount for a surveillance Compares results of 2 ‘independent’ reporting systems Calculates number of cases missed by both systems Estimated total number of cases derived Determines reporting completeness for a surveillance
system Assumptions:
Closed population No loss of tags Simple randomness Independency
Methods
Linked IPD surveillance data with Hospital Inpatient Discharge Data (HIDD) by deterministic data linkage
Identified potential IPD cases in HIDD by ICD-9 codes ICD-9 Codes Definitions
320.1 Meningitis due to S.pneumoniae
038.2 Septicemia due to S. pneumoniae
481 Pneumonia due to S. pneumoniae
320.2 Streptococcal meningitis
041.2 S. pneumoniae as the cause of bacterial infectionclassified elsewhere and of unspecified site
Specific
Nonspecific
Data Linkage Results
HIDD n=1,287
Surveillance n=1,191 (~67% initially passive)
Data Linkage Results
HIDD n=1,287
Surveillance n=1,191 (~67% initially passive)
Linked n=523
Data Linkage Results
HIDD n=1,287
Surveillance n=1,191 (~67% initially passive)
Surveillance only n=668 (~79% hospitalized)
HIDD only n=764
Linked n=523
Data Linkage Results
HIDD n=1,287
Surveillance n=1,191 (~67% initially passive)
Surveillance only n=668 (79% hospitalized)
HIDD only n=764
Linked n=523
Data Linkage Results
IPD-specific codes n=62
Nonspecific codes n=702
HIDD n=1,287
Surveillance n=1,191 (~67% initially passive)
Surveillance only n=668 (79% hospitalized)
HIDD only n=764
Linked n=523
Data Linkage Results
Census approach 62 (100%) Reviewed
IPD-specific codes n=62
Nonspecific codes n=702
HIDD n=1,287
Surveillance n=1,191 (~67% initially passive)
Surveillance only n=668 (79% hospitalized)
HIDD only n=764
Linked n=523
Data Linkage Results
Census approach 62 (100%) Reviewed 4 confirmed cases
IPD-specific codes n=62
Nonspecific codes n=702
HIDD n=1,287
Surveillance n=1,191 (~67% initially passive)
Surveillance only n=668 (79% hospitalized)
HIDD only n=764
Linked n=523
IPD casesn=4
Not casesN=58
Data Linkage Results
IPD-specific codes n=62
Nonspecific codes n=702
102 (15%) Systematic sampling102 (100%) Reviewed4 were confirmed IPD cases
HIDD n=1,287
Surveillance n=1,191 (~67% initially passive)
Surveillance only n=668 (79% hospitalized)
HIDD only n=764
Linked n=523
Sample requestedn=102
Data Linkage Results
IPD-specific codes n=62
Nonspecific codes n=702
Systematic sampling102 (100%) Reviewed4 were confirmed IPD cases
HIDD n=1,287
Surveillance n=1,191 (~67% initially passive)
Surveillance only n=668 (79% hospitalized)
HIDD only n=764
Linked n=523
Estimated IPD casesn=28 95%CI (8-68)
Estimated Not casesN=674
Sample requestedn=102
Data Linkage Results
Census approach 62 (100%) Reviewed 4 confirmed cases
IPD-specific codes n=62
Nonspecific codes n=702
HIDD n=1,287
Surveillance n=1,191 (~67% initially passive)
Surveillance only n=668 (79% hospitalized)
HIDD only n=764
Linked n=523
IPD casesn=4
Not casesN=58
Systematic sampling102 (100%) Reviewed4 confirmed IPD cases
Estimated IPD casesn=28 95%CI (8-68)
Sample requestedn=102
Estimated Not casesN=674
Final Capture-Recapture Results
IPD Surveillance System Sensitivity : 1,191/1,264 = 94%*
HIDD IPD Sensitivity : 555/1264 : 44%*
* 95% confidence interval estimate pending further review/validation
Identified by HIDDYes No Total
Detected by Surveillance System
Yes 523 668 1,191No 32 (12-72) * 41Total 555 1,264*
ICD-9 Code Distribution by IPD Case Status within HIDD post Laboratory Reports Review
IPD Not IPD TotalIPD-specific codes 272 58 330
Nonspecific codes 283 674 957
Total 555 732 1,287
SEN = 49%PVP = 82%
SEN = 51%PVP = 30%
Missed Cases by Culture Site
HIDD identified Cases 1-6, not previously identified by IPD surveillance
Case Culture Site Reason Missed
1 Blood Failure of lab to code all sterile site isolates as invasive
2 Blood Infection control practitioner generated list of cases
3 Blood Unknown
4 Blood Unknown
5 Blood Unknown
6 Body fluid Lab misnomer of body fluid; actually synovial fluid
7 Body fluid Initially considered not a case by NMDOHReclassified: body fluid was actually pleural fluid/empyema
8 Abscess aspirate Initially considered not a case by NMDOHReclassified: aspirate was from sternoclavicular joint abscess
n=4 (2.564%)n=18 (11.54%)
n=28 (17.95%)
n=43 (27.56%)
n=63 (40.38%)
No Micro Non IPD(Others/Unclassified)Non S. Pneum. Isolated Negative MicroS. Pneum./non Sterile Site
Combined Specific and Nonspecific SampleExcluded Hospital Admissions by Final Status
Limitations
Sampling instead of census approach Sampling variability Small sample size Low precision Incomplete sampling frame
Time period chosen HIDD data only included hospital admissions Systems were not entirely independent , potentials for:
Positive dependency phenomenon Underestimation of total IPD cases Overestimation of IPD surveillance system sensitivity
Strengths
Accurate diagnosis of IPD Correct identification of IPD cases Closed population Deterministic data linkage reduces false matches
Manually reviewing all the linked data and correcting for all the identified false matches
Systematic sampling
Conclusions
High NM IPD surveillance sensitivity ABCs
A sample of hospitalization data yielded eight additional IPD cases HIDD IPD sensitivity? ICD-code dependent
Recommendations In New Mexico,
Periodic review of HIDD data may be worthwhile. This identified additional IPD cases but required a lot of work
A study of the hospitalized IPD cases yet unidentified by HIDD is warranted
States relying on passive reporting without resources to do active surveillance might use IPD-specific ICD-9 codes to improve IPD surveillance
IPD case-ascertainment deficiencies, including hospitalization coding problems, should be addressed through coding study
Capture-Recapture methods may be used to improve surveillance case findings
For more information please contact Centers for Disease Control and Prevention
1600 Clifton Road NE, Atlanta, GA 30333Telephone, 1-800-CDC-INFO (232-4636)/TTY: 1-888-232-6348E-mail: [email protected] Web: www.cdc.gov
The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.
Acknowledgments
Office of Surveillance, Epidemiology, and Laboratory ServicesScientific Education and Professional Development Program Office
NM DOH: Michael Landen (co-author) Joseph Bareta, Joan Baumbach, Camille Clifford, Paul Ettestad, Jessica Jungk, Megin Nichols, Terry Reusser, Mack Sewell, Chad Smelser, Brian Woods
CDC: Julie MagriDiana Bensyl, Betsy Gunnels, Sheryl Lyss
n=22 (37.93%)
n=20 (34.48%)
n=2 (3.448%)
n=11 (18.97%)
n=3 (5.172%)
n=41 (41.84%)
n=8 (8.163%)n=2 (2.041%)
n=32 (32.65%)
n=15 (15.31%)
Specific Nonspecific
S. Pneum./non Sterile Site Non S. Pneum. IsolatedNo Micro Negative MicroNon IPD(Others/Unclassified)
Graphs by IPD ICD-9 Codes
Excluded Hospital Admissions by Final Status
Calculation of Completeness of Reporting by the Two-Source Capture-Recapture Method
Source 2 casesSource 1 cases
TotalReported Not reported
Reported C N2 SNot reported N1 XTotal R N
C = number of people identified by both sourcesN2 = number of people identified only in data source 2S = number of people identified in data source 2N1 = number of people identified only in data source 1R = number of people identified in data source 1X = number of cases not reported to either system(estimated)N = estimate of total number of cases
......................................................N = RS/CCompleteness of source 1 = R/NCompleteness of source 2 = S/N
......................................................Var (N) = ( R * S * N1 * N2 ) / C3
95% CI = N ± 1.96 Var (N)1/2
ICD-9 Distribution within HIDDPreliminary Analysis
ICD-9 codes Linked (Reported)
Nonlinked (Unreported)
Total
IPD-Specific 317 104 421
Nonspecific 241 625 866
Total 558 729 1287
Prevalence Rate Ratio ~ 2.6
ICD-9 Distribution within HIDD
75%
25%
IPD-specific
Linked(Reported)
NonLinked(Unreported)
28%
72%
Nonspecific
Linked(Reported)
NonLinked(Unreported)
57%
43%
Linked(Reported)
IPD-specificNonspecific
14%
86%
NonLinked(Unreported)
IPD-specificNonspecific
Some Reasons for Misclassification of HIDD IPD Cases
Keypunch Coding error Abstraction error Physician error (Rule out IPD) Physician error (other) No error; clinically compatible
Linkage Lessons Sequential deterministic linkage Overall rate of false +ve matches: 5.97% Overall rate of false -ve matches: 0.41%
Recommendations to Improve IPD Surveillance
Direct electronic reporting of laboratory data Identification of missed opportunities for reporting System to automatically remind treating doctors Provision of updatable computer software Hospital coders to seek evidence of documented reporting Audit of selected laboratories Studies to identify coding issues and reasons for under
reporting
Demo
Source 1 yes noyes 6 9 15no 1 2
7 18
Scenario 1:Source 2
Slides Master1-Title2-Invasive Pneumococcal Disease (IPD)3-Importance of IPD Surveillance Systems4-IPD Surveillance in New Mexico5-Questions/Objectives6-Capture-Recapture Method7-Methods/ ICD codes(8-16) Data Linkage Results17-Full flow diagram18-Final Cap-Recap Results19-ICD-9 Codes by IPD Case Status20-Missed Cases by Culture Site21-Excluded Hospital Admissions by Final Status22-Limitations23-Strengths24-Conclusions25-Recommendations26-Acknowledgments27-Empty28-Excluded Hospital Admissions by Final Status by ICD-9 codes29-Cap-Recap Calculus30-ICD-9 Distribution: Preliminary analysis31-ICD-9 Distribution: Pie Charts 32-Some Reasons for HIDD IPD Cases Misclassifications33-Linkage Lessons34-Recommendations to Improve IPD Surveillance35-Cap-Recap Sampling Demo36-Cap-Recap IPD Assumptions37-Scenarios38-Slide Master