The sensitivity and timeliness of reporting hospitalized cases of notifiable diseases to CEDRS differed across the eight diseases examined and according to the methodology used to identify true hospitalized cases. Reliance on ICD-9-CM discharge diagnoses from the IHD database worked better for certain notifiable diseases than for others. Review of medical records to confirm discharge diagnoses resulted in improved sensitivity and timeliness of reporting. Hence, medical record abstraction should be considered a necessary component of evaluating notifiable disease surveillance systems when using hospital discharge databases.
Our study demonstrated that the utility of using discharge diagnoses is dependent upon the complexity of the notifiable disease case definition and a clinician's ability to make a diagnosis based on clinical signs and symptoms alone. Using discharge diagnoses to identify true cases of notifiable diseases was most useful for diseases (i.e., salmonellosis, shigellosis, and legionellosis) that had simple case definitions and nonspecific clinical signs (e.g., diarrhea, vomiting, or chest oscillation suggestive of pneumonia) that would require laboratory confirmation of an etiologic agent before diagnosis. In contrast, discharge diagnoses were least useful for diseases (i.e., H. influenzae invasive disease and pertussis) that had complex case definitions or unique clinical signs (e.g., cough paroxysms) that might lead to a presumptive diagnosis in the absence of laboratory confirmation.
Both N. meningitidis
invasive disease and H. influenzae
invasive disease are included in Colorado's active laboratory-based surveillance (i.e., ABCs) and were therefore expected to have high sensitivity.6
By using the discharge diagnosis method only, we determined that the sensitivity of N. meningitidis
invasive disease was 90% and H. influenzae
invasive disease was 17%. To meet the CSTE/CDC definition for a confirmed case, both conditions require isolation of the bacterium from a normally sterile site (e.g., blood or cerebrospinal fluid). For H. influenzae
invasive disease, only 15% of the medical records reviewed were for confirmed cases; the majority of unconfirmed cases had H. influenzae
isolated from a nonsterile site (i.e., sputum) and a diagnosis of pneumonia. The removal of these noninvasive cases during the medical record review substantially improved the reporting sensitivity of H. influenzae
invasive disease from 17% to 100%.
For pertussis, one-third of IHD cases that underwent medical record review were not classified as confirmed based upon the available information. The majority of unconfirmed cases met the clinical description in the CSTE/CDC case definition but had a missing or negative laboratory result in the medical record (i.e., they were probable or epidemiologically linked cases). This finding indicates that clinicians might be more likely to diagnose pertussis without laboratory confirmation either because of confidence in recognition of clinical signs or the low predictive value of the diagnostic tests. Limiting the population to confirmed cases through medical record review improved reporting sensitivity of pertussis from 59% to 100%. We did not assess the sensitivity of reporting probable or epidemiologically linked cases in CEDRS.
Medical record review had less of an impact on the reporting of legionellosis cases. The identification of one additional case in CEDRS by using information available from the medical record increased the reporting sensitivity of legionellosis from 80% to 90%.
Hepatitis A had suboptimal sensitivity and timeliness according to both the discharge diagnosis and medical record review methods. A limited number of cases of hepatitis A were confirmed during medical record review; unconfirmed cases usually had a prior history of hepatitis A infection indicated in the patient's medical record. After medical record review, the reporting sensitivity of hepatitis A improved from low (21%) to moderate (67%), but timeliness remained inadequate. These findings might reflect unstable estimates resulting from the limited sample size (n=6) or the need for improved hepatitis A reporting within the TCHD jurisdiction.
Challenges to using the discharge diagnosis method alone for surveillance evaluations can result in an underestimation of sensitivity and timeliness. First, IHD databases are designed for administrative and billing purposes, and inconsistencies exist in assigning and ordering ICD-9-CM codes in the discharge diagnoses fields across hospitals and physicians. Second, the denominator of hospitalized cases obtained from the IHD database might be overestimated for multiple reasons: (1) a notifiable disease discharge diagnosis might not meet the criteria for a confirmed case according to the CSTE/CDC surveillance case definitions; (2) a notifiable disease discharge diagnosis listed as the second or third diagnosis can indicate past rather than current illness; (3) the IHD database might contain multiple records per patient, and identifying and removing duplicate records (i.e., hospital transfers or readmissions) is difficult without personally identifying information; and (4) the IHD database might contain records of patients who reside outside TCHD's jurisdiction, because records were selected according to residential ZIP code rather than street addresses.
Third, the lack of personally identifying information in the IHD database limited the ability to merge records with the CEDRS database because: (1) multiple patients with the same 12-digit identifier might be included in one or both databases, (2) the same patient with incorrect information in one or both databases is unlikely to match, and (3) lack of patient name and street address makes verifying the accuracy of the matches difficult. Lastly, use of admission date as a proxy for diagnosis date in the discharge diagnosis method likely increased the calculated report time. This is especially true for diseases that require more time for laboratory confirmation (e.g., culture results for H. influenzae).
Review of medical records provided the clinical and demographic information necessary to overcome the challenges of the discharge diagnosis method described previously, resulting in improved sensitivity and timeliness of reporting for each of the four diseases examined. Specifically, we limited the denominator of hospitalized cases to confirmed cases, according to the national CSTE/CDC case definition. We then identified and removed data regarding patients who had been transferred or readmitted, or who resided outside TCHD's jurisdiction. Matches between CEDRS and IHD databases were verified by using patients' names, and laboratory result dates were available for the majority of cases. However, the medical record review method has limitations that deserve mentioning.
First, TCHD did not obtain identifiers for medical record review and had to partner with the state health department to obtain medical record numbers from the IHD database. Second, requesting medical records is time and resource intensive, as is traveling to different hospitals to conduct chart abstraction. This process can be simplified if electronic medical records are available to the state health department, as was the case with one hospital in this study. Third, we suspected that using specimen collection date plus one day as a proxy for laboratory result date increased the calculated report time. However, a post-hoc analysis determined that using collection date plus two days made no difference for three of the four diseases but improved the timeliness of hepatitis A reporting from 25% to 50%. Lastly, limited sample size can affect the calculation of sensitivity and timeliness for notifiable diseases that have a low incidence or do not often result in hospitalization, as was the case for hepatitis A and H. influenzae invasive disease. For such diseases, the medical record review method should only be considered for health departments that serve a substantial population or are able to combine multiple years of surveillance data.