Validation of communicable disease reporting from hospitals using the hospital discharge database, Arizona, 2007 2009 (Poster is shared here as an 8.5 x11 document for easier viewing. All content is identical.) Laura M. Erhart, MPH laura.erhart@azdhs.gov Arizona Department of Health Services Background: Public health communicable disease reporting is critical to identifying outbreaks and controlling disease spread. Most states rely on passive disease surveillance for identifying cases of nationally notifiable disease. In Arizona, selected diseases are reportable by healthcare providers, as specified in the Arizona Administrative Code (AAC) R9 6 202. Positive laboratory tests for certain morbidities are also reportable by laboratories (AAC R9 6 204). Periodic evaluation of the surveillance system is important for identifying reporting gaps and strengthening the system. The Arizona Department of Health Services (ADHS), in partnership with county health departments, conducted a three year surveillance validation study with the following goals: Assess the completeness of reporting of hospitalized or emergency department cases, for selected morbidities; Examine differences in reporting across morbidities, reporting categories, and years; and Refine validation methods for future use. Methods: Morbidities Included: Morbidities included in the analysis met the following criteria: a) provider reportable in Arizona, and b) specific ICD 9 code(s) could be identified. Each condition was also categorized according to its reporting timeframe. The 39 morbidities included are listed in Table 1. Data Sources: The Arizona Hospital Discharge Database (HDD) was used to identify communicable diseases diagnosed at acute care hospitals in Arizona; the criteria differed slightly by year (see Table 2). Arizona s communicable disease surveillance database (MEDSIS) was used to identify cases reported to the public health system in 2007 through 2009. Cases with residency outside of Arizona were excluded in 2007 and 2008. Reported cases later ruled out for not meeting the Erhart 1
public health case definition were included. MEDSIS includes cases from both providers and laboratories. Record matching: A unique identifier was created for all selected records in each database, based on date of birth, sex, and part of the first and last names. Records were matched on an exact match of this unique identifier, within the defined year (e.g., 2007 HDD records were matched with 2007 MEDSIS records). Unmatched records from selected facilities were also manually scanned for additional matches. Record review: County health departments reviewed medical records for a sample of unmatched HDD records to identify cases that should have been reported. Analysis: The following were calculated: 1. Number of HDD records identified, statewide, for each morbidity and reportable category, and the number and proportion reported in MEDSIS. 2. Number of medical records reviewed for each morbidity and number and percentage of reviewed HDD records that should have been reported. 3. Revised estimates of the proportion of reportable cases in HDD that were reported and recorded in MEDSIS, based on the chart review findings: Revised percentage = number matched / (number matched + (number unmatched x percentage of reviewed charts that should have been reported)) Table 1. Conditions included for analysis Report within 24 hours (16) Report within 1 working day (8) Report within 5 days (15) Anthrax Plague Brucellosis Amebiasis Hepatitis A Botulism Poliomyelitis Cholera Aseptic Legionellosis Meningitis: Viral Diptheria Rabies in a human Encephalitis, viral or parasitic Campylobacteri osis Malaria Hemolytic uremic syndrome Severe acute respiratory syndrome Mumps Cryptosporidiosi s Rocky Mountain spotted fever Listeriosis Smalllpox Q Fever Dengue Salmonellosis Measles Tularemia Rubella (German Giardiasis Shigellosis (rubeola) measles) Meningococcal Typhoid fever Rubella syndrome, Haemophilus West Nile virus Erhart 2
invasive disease congenital influenzae: invasive disease Pertussis Yellow Fever Typhus Fever Hantavirus (whooping infection cough) infection Table 2. HDD records included in analysis 2007 2008 2009 Dates Used Full year January June Full year Facilities Used 15 All All Records Used Inpatient* Inpatient & Emergency Department Number of Diagnosis Codes used for each record to identify reportable condition Diagnosis 1 through 9 Diagnosis 1 through 25 Inpatient & Emergency Department Diagnosis 1 and 2 *Prior to 2008, the reporting of patient names was not required for Emergency Department records. Results: 1,974 records were included in the analysis (1,685 inpatient records and 289 emergency department records), for the selected diagnoses from hospitals statewide. These include: 2007: 454 inpatient records from the 15 hospitals selected for validation 2008 (January June): 580 records (496 inpatient, 84 emergency department) 2009: 940 records (735 inpatient, 205 emergency department) Table 3. Number of reportable morbidities, by reporting category, identified in HDD and percent matched to surveillance data REPORTING CATEGORY TOTAL IN HDD TOTAL MATCHED TO PERCENTAGE MATCHED MEDSIS 24 HOUR and 1 WORKING DAY REPORTABLES 126 44 35% 5 DAY REPORTABLES 1,816 644 35% TOTAL 1974 696 35% No records were identified in the three years for: botulism, brucellosis, hantavirus, measles, mumps, poliomyelitis, Q fever, rabies, Rocky Mountain spotted fever, rubella syndrome (congenital), smallpox, typhus fever, or yellow fever. Erhart 3
Table 4. Number of reportable morbidities, identified in HDD and percent matched to surveillance data MORBIDITY TOTAL IN HDD TOTAL MATCHED TO PERCENTAGE MATCHED MEDSIS Listeriosis 2 2 100% Cholera 1 1 100% Plague 1 1 100% Tularemia 1 1 100% Salmonellosis 296 261 88% Legionellosis 18 13 72% Typhoid fever 10 7 70% Meningococcal invasive disease 29 20 69% Malaria 37 23 62% Pertussis * 21 10 48% Shigellosis 12 5 42% Haemophilus influenzae, invasive disease 232 69 30% Aseptic Meningitis, Encephalitis (viral), West Nile Virus * 876 204 23% Hepatitis A 315 55 17% Diphtheria * 38 2 5% Hemolytic Uremic Syndrome * 12 0 0% Anthrax 5 0 0% Rubella 5 0 0% SARS * 1 0 0% * Morbidities for which laboratory reporting alone cannot capture all reportable cases (no laboratory test available, or cases may rely on a clinical diagnosis). Table 5. Number of reportable morbidities identified in HDD and matched to surveillance data, by year YEAR TOTAL IN HDD TOTAL MATCHED TO PERCENTAGE MATCHED MEDSIS 2007 454 179 39% 2008 580 143 25% 2009 940 374 40% Erhart 4
Table 6. Number of unmatched records reviewed, and revised proportion reported REPORTING CATEGORY TOTAL CHARTS NUMBER OF CHARTS THAT SHOULD HAVE BEEN REPORTED REVISED ESTIMATED OF PERCENT OF HDD CASES REPORTED 24 HOUR and 1 WORKING DAY REPORTABLES 28 9 73% 5 DAY REPORTABLES 287 86 59% Table 7. Number of unmatched records reviewed, and revised proportion reported, for selected morbidities MORBIDITY TOTAL CHARTS NUMBER OF CHARTS THAT SHOULD HAVE BEEN REPORTED REVISED ESTIMATE OF PERCENT OF HDD CASES REPORTED Anthrax 3 0 Hemolytic Uremic Syndrome 10 4 0% Meningococcal invasive disease 4 2 82% Pertussis 5 3 60% Aseptic Meningitis, Encephalitis (viral), West Nile Virus 85 46 36% Haemophilus influenzae, invasive disease 39 3 85% Hepatitis A 120 14 64% Salmonellosis 19 10 93% Shigellosis 2 2 42% Reasons why some unreported records were not considered cases (and were appropriately not reported): Anthrax: no mention of morbidity & symptoms were not consistent (1); data entry error, diagnosed with pinworms (1) Giardiasis: confusion with GERD (2) Haemophilus influenzae: patient did not have invasive disease (most common reason) Hepatitis A: history of hepatitis A (most common reason) Several morbidities: no clinical or laboratory indications of the morbidity in question Erhart 5
Conclusions: Initial analysis of the percentage of HDD records reported indicated low reporting in each of the three years, and across all three reporting category, including among the 24 hour reportables, a group intended to include diseases needing urgent public health intervention. Record review of unreported cases revealed that many of these records were not true public health cases. In 2009, we changed to using only primary and secondary diagnosis codes to pull records from the HDD, in order to improve the specificity of the analysis. In addition to improving the initial percentage reported, a higher proportion of the unreported cases reviewed were true cases, suggesting better specificity. The coding of several diagnoses in the HDD was not always reliable. This is especially concerning for several morbidities highly important for public health officials, such as anthrax and SARS. Several morbidities contributed substantially to the low initial proportions reported: o 24 hour reportables: Diphtheria: Chart review for three records did not indicate diphtheria. For the 2009 analysis we eliminated one of the ICD 9 codes used to identify diphtheria This code was too broad to be used effectively and accounts for most of the diphtheria records listed here. Pertussis: Almost half of the cases reviewed should have been reported. As pertussis can be diagnosed without laboratory findings, missed cases may indicate lack of provider reporting. o 5 day reportables: Aseptic meningitis: Chart review indicated 60% of unmatched records should have been reported. Aseptic meningitis is only reportable by providers, not laboratories. Hepatitis A: Chart review indicated only 12% of the unmatched records needed to be reported; many appear to be coded based on past medical history. Haemophilus influenzae infections: Only 5% of reviewed records needed to be reported. This represents a mismatch between the ICD 9 code, which is for all H. influenzae infections, and the public health reporting of only invasive infections. Limitations: Reporting by providers cannot be distinguished from reporting by laboratories. While we are assessing the reporting of patients seen in hospitals, laboratory reporting may contribute substantially for many morbidities. HDD is intended for administrative and billing purposes, not surveillance, and is not available rapidly enough to identify cases for public health intervention. Erhart 6
Public health case definitions and ICD 9 codes match better for some morbidities than others. Recommendations: This type of analysis is useful for estimating the proportion of inpatient or emergency department patients reported. Record review of at least a portion of unreported records is critical for revising reporting percentages and understanding why some records may not be have been reported. To improve the quality of the reporting estimates without extensive chart review, limit the ICD 9 codes used to those that are most specific for the reportable condition. While there may be additional public health cases included in some broader categories, specificity is important in this case for identifying the correct denominators. If the objective is instead to identify all cases of public health importance, sensitivity may be more important and the diagnostic code list should be expanded. Limit the record selection to only the primary and secondary diagnoses within each record. Finally, this analysis served as a useful tool for health departments to interact with infection preventionists to reinforce reporting protocols when unreported cases were identified. Special thanks to members of the Surveillance Validation Work Group consisting of state and local health department epidemiologists Erhart 7