|
BACKGROUND/RATIONALE:
Early illness detection is essential to minimize morbidity and mortality in the event of infectious disease outbreaks of public health significance e.g. a severe form of influenza, SARS or bioterrorism. The automated monitoring of electronic health information offers the potential to enhance disease surveillance compared to systems that rely mostly on manual case reporting. The availability, selection and arrangement of electronically extractable data elements affect an automated system's ability to measure applicable health-related events at the earliest stage. Today's most sophisticated systems aim at identifying disease at a pre-diagnostic stage, and thus are challenged by the difficulty of detecting an outbreak signal among the "noise" of common syndromes such as an influenza-like illness (ILI) or diarrhea. These systems are further limited to a narrow range of electronic data types, often obtained from unrelated sources. A system that could meaningfully relate a broad array of clinical data around unique individuals could follow the progress of each patient over time and thus identify aberrancies not only in the incidence, but also in the course their illness.
OBJECTIVE(S):
Our overall goal is to automate the use electronic clinical data from CPRS to enhance efficiency, timeliness, and validity of outbreak detection. We hypothesize that expanding the scope of an automated surveillance system to include illness progression and severity will shorten the time it takes to recognize disease outbreaks that pose a significant threat to veterans and public health.
Our specific aims are to: 1) refine and validate automated methods to interpret acute respiratory infections (ARI) progression and severity; 2) develop and validate statistical signals for surveillance that combine disease incidence and severity; 3) use simulated outbreaks of ARI to evaluate the sensitivity and specificity of statistical models that account for disease severity compared to models that do not; 4) systematically vary simulated outbreak characteristics so as to optimize the automated surveillance methods developed in Aims 1 and 2 for timely, sensitive and specific outbreak detection; 5) initiate Public Health Information Network (PHIN)-compliant, real-time electronic transmission of health record data to support public health disease surveillance.
METHODS:
Clinical data are extracted from VA's Computerized Patient Record System (CPRS) and transferred to relational tables on a SQL Server. To build predictive models, relevant data are grouped and classified along different dimensions of respiratory disease severity: diagnostic and procedures codes, type of health care delivery, therapeutic and monitoring efforts, physiological, laboratory and imaging results, and the free text of clinical notes and XR reports. CDAs are created using iterative regression and tested against a reference manual structured chart review for their ability to identify ARI and to follow disease severity over time at the single-patient level. We recreated historical background case count time series by applying the most successful CDAs to past EMR data. We injected factitious influenza cases to CDA-specific backgrounds using an age-structured modeled influenza epidemic and then used a modified CUSUM statistic daily for 80 days to detect the outbreak. This injection/prospective-surveillance cycle was repeated each week of the study year. To distinguish between true- and background-positive alarms, the daily statistics were performed on paired background+injection vs. background-only time series. We computed two whole-system benchmarks: 1) the average "Detection Delay", from the time of each injection to the first true-positive alarm; 2) the "Workload", defined as the yearly number of cases included in all the background-positive alarms.
FINDINGS/RESULTS:
We found that seven (7) percent of ARI patients have a fever, yet 70% of reported influenza patients have a fever. Thus, monitoring for Febrile-ILI ignores 30% of patients with influenza who do not have a fever (reduces case-finding sensitivity), but also eliminates 93% of the ARI background noise while enriching the epidemic signal for real influenza patients (increases specificity and signal-to-noise ratio). In a first step toward evaluating the benefit of incorporating measures of disease severity into a surveillance system, we compared the "Detection Delay" and the "Workload" whole-system benchmarks for simulated surveillance systems optimized to target either "any ARI", or only that subset of ARI patients who were also febrile when encountered (ILI). Compared to the best ICD-9-only CDAs optimized for the VA, ARI CDAs that also included information about cough remedies and a computerized text analysis for at least two non-negated ARI symptoms (via an adapted NegEx algorithm) decreased Detection Delay from 40 to 30 days, but increased Workload from 267 to 483 cases/year. The best (febrile) ILI-targeted CDA further reduced Delay to 22 days and Workload to 121 cases/year.
IMPACT:
Our data to date suggest that automated surveillance systems for influenza should evolve to integrate information from prescriptions and from the free text of a clinical note. Case detection methods that take advantage of information from the full EMR and that focus only on those ILI cases that are febrile can lower both the delay and the workload required to detect an influenza epidemic in the community.
PUBLICATIONS:
None at this time.
DRA:
Health Services and Systems
DRE:
Epidemiology
Keywords:
Research method, Bioterrorism
MeSH Terms:
none
|
