To evaluate the sensitivity of the surveillance system, personnel at all surveillance sites conducted a retrospective review of death records from their surveillance areas, and three sites (California, Connecticut, and Oregon) reviewed all hospital discharge data in their areas for a period of at least 1 year. All death certificates for the age groups included in the surveillance were reviewed for the presence of specific International Classification of Disease codes (ICD-9), selected for their potential to identify unexplained deaths due to possibly infectious causes (3). Persons whose death records included ICD-9 codes indicating a disqualifying underlying medical condition were excluded. Once potential cases were identified, the patients’ medical records were reviewed. If the records were not available, the primary physician was contacted to determine if the patient met the surveillance case definition. The sensitivity of the surveillance system for detecting deaths (S_D) was calculated by dividing the number of deaths (D1) detected through surveillance alone by the total number of deaths (D1+D2) detected through both surveillance and death record review (D2): S_D=D1/D1+D2. The sensitivity of the surveillance system (S_C) for detecting critical illness cases was calculated by dividing the number of such cases (C1) detected through surveillance alone by the total number of cases (C1+C2) found through both surveillance and hospital discharge review (C2): S=C1/C1+C2.

For patients meeting the case definition, surveillance site personnel completed a case report form that included demographic, epidemiologic, and clinical information. This information was collected through interview of physicians caring for the patient, review of the medical record, and contact with the patient or the patient’s family. Cases were assigned a clinical syndrome depending on the predominant system involved, on the basis of information provided by the physician. These syndromes included neurologic (encephalitis, meningitis), cardiac (myocarditis, pericarditis, endocarditis), respiratory (pneumonitis), and hepatic (hepatitis). Syndromes such as sepsis, in which no predominant organ system was involved, were classified as “other.” The hospital laboratories were requested to save all remaining clinical specimens obtained as part of routine clinical management, including biopsies and autopsies.

For the first 2 years of the study, the project investigators selected diagnostic tests individually for each case. Decisions were made on the basis of clinical, epidemiologic, and histologic data; previous laboratory testing ordered by the health-care providers; and availability, timing, quality, and quantity of clinical specimens. In the third year of the project, based on information gained to date, a set of standardized syndrome-specific laboratory testing protocols was developed for respiratory, neurologic, cardiac, and hepatic syndromes (Appendix II).These protocols prioritized testing based on available clinical and epidemiologic information and a differential diagnosis; they guided a first round of laboratory testing which, if negative, prompted a customized second round of testing. Cases that did not fit any of these four syndromes were discussed by the project investigators on an individual basis.

Cases were defined as having definite, probable, possible, or no microbial etiology (Table 1). These levels of certainty for the causal role of an infectious agent reflected the integration of several factors, including the relationship of anatomic site of detection to site of disease, reliability of the method, and whether the putative agent was a known cause of the clinical syndrome under investigation. Cases were classified as explained if results showed a definite or probable disease cause and as unexplained if results indicated a possible infectious cause or none at all.

Analysis was performed by SAS 6.12 (SAS Institute, Cary, NC). Denominators for the population under surveillance, obtained from the 1992 intercensus (14), included all persons in the age groups under surveillance at the various sites; denominators including only previously healthy persons are not available, and no attempt was made to estimate this fraction. Data from the surveillance population were standardized to the U.S. population by race and age to project the number of cases occurring nationally. The chi-square test was used to compare the distribution of characteristics between explained and unexplained cases. A p value ≤0.05 was considered statistically significant.

This study is the first to measure the population burden of unexplained deaths and critical illness from possibly infectious causes in the United States. To our knowledge, this is the first public health attempt to describe the features of this problem, in spite of its clinical complexities. This project established the infrastructure needed to detect UNEX cases, attempt to identify their etiology, and ultimately identify new infectious agents. However, since this project was a pilot study, it was difficult to standardize many of its aspects. Many lessons were learned during this project, whether related to the best surveillance methods to use or the laboratory testing process. In addition, data obtained in the first 3.5 years of this project suggest that UNEX occur in previously healthy persons at rates similar to those of other conditions of clear public health concern and priority (18). Of obvious concern is also the large proportion of these deaths and severe illnesses that remains unexplained after extensive laboratory testing. Our findings highlight the substantial limitations of available diagnostic tests for infectious diseases and the need for improved tests and novel approaches to identify infectious disease agents.

Our surveillance estimated the burden of disease only among previously healthy persons 1 to 49 years of age. Since a different age cut-off was used in Oregon, the final rates of disease were adjusted for age and race. The lower age limit was chosen to avoid confusion with congenital problems seen in infants but to include most children in day care, where infectious diseases are common and new infectious diseases might spread rapidly. The upper age limit was intended to exclude an expected increased proportion of unexplained deaths due to noninfectious causes among persons ≥50 years of age. Although immunocompromised patients are more susceptible to a variety of infectious diseases, available resources and a concern that the clinical relevance of novel microbial findings would be more difficult to interpret in immunocompromised persons compelled us to focus on previously healthy persons. In addition, many of the new infectious diseases first identified in these persons have subsequently been found to affect persons with normal immune systems (19,20).

The surveillance methods adopted during this project were customized to meet the objectives of this study, taking into consideration the limitations of local resources; therefore UNEX cannot be easily compared with other classical surveillance systems. The different methods of surveillance used at the four sites allowed us, through the surveillance audits and validation, to determine how these differences affected case-finding. For example, investigators in Connecticut were able to detect most UNEX cases largely because they conducted more active surveillance in a smaller population base; in this site, surveillance focused on all seven hospitals in New Haven County. At the academic tertiary-care hospital, EIP staff reviewed ICU admission logs and communicated with clinicians daily. At the other six hospitals, a stimulated passive surveillance system was used in which physicians and infection control practitioners were given reminders several times per year. The active prospective method captured a greater proportion of total cases (86% of cases at the single hospital) then did the passive methods (50% of total cases at the six remaining hospitals).

If this surveillance is to be expanded, different methods may be chosen, depending on availability of resources and overall objectives. Less resource-intensive passive surveillance may be used if the goal is to monitor trends in disease occurrence. For example, although analyzing all death certificates for UNEX cases would be prohibitively time-consuming, electronically searching only the certificates in which the manner of death was recorded as natural, undetermined, or pending investigation could substantially decrease the workload. Under a passive system, maintaining good communication between study staff and clinical staff (clinicians, pathologists, and infection control practitioners) is critical and aided by the provision of diagnostic testing not locally available (such as serologic testing for hantavirus or toxin testing for botulism) and timely feedback of study results. Such collaboration may be critical to early diagnosis of diseases that produce characteristic clinical syndromes (e.g., potential bioterrorist agents such as botulism) or that are not readily confirmed by clinical laboratories.

Before initiating this project, we had reviewed multiple cause-of-death data for the United States to estimate the number of unexplained deaths from possibly infectious causes at these EIP sites (3). In 1992, the rate of unexplained deaths among healthy persons 1 to 49 years of age was 8.9 per 100,000 population. The discrepancy between this rate and that found in our study (0.5 per 100,000) is likely due to the low specificity of ICD-9 codes in excluding persons with previous health problems, as well as the problems related to retrospective analysis in general. For at least two reasons, we expect that the incidence of UNEX found in this study represents only a minimal estimate of the true burden of this problem. First, the denominator in our calculations included all persons in our designated groups, since we chose not to estimate the fraction of previously healthy persons in the surveillance populations at the four sites. Second, the differences in incidence rates between the four surveillance sites and results of the surveillance audits support the assumption that the overall rate detected was a minimal estimate of overall disease.

An important unresolved issue from our study is the large proportion of cases that remained unexplained, even after extensive laboratory testing. Although a standardized protocol for testing was used only during 1998, the proportion of explained cases before and after this protocol was used did not differ substantially. Some illnesses may have noninfectious causes, especially given the lack of specificity in our clinical criteria for case inclusion and in the features of infection in general. In cardiac syndromes, for example, myocarditis and myocardial infarction can have very similar presentations. Some cases may have been caused by microbial products such as toxins without the presence of the organism or substantial amounts of its nucleic acids. Laboratory methods for screening and detection of toxins remain inadequate. For some patients, specimens were not available from the primary site of disease, were severely limited in quantity, or were only available from late in the course of the disease; in many cases, multiple serum specimens were not available, autopsies were incomplete, and tissue specimens were obtained only from unaffected organs. Finally, the breadth of our testing methods may not have been adequate. Since broad-range PCR methods were applied only to bacteria and a limited range of viruses, many other potential agents may have been missed. Our approach to the detection of viral pathogens relied more heavily on serologic and immunohistochemical techniques, in part because of the difficulty in designing a comprehensive set of consensus PCR primers for all known viral families (21). In our study, viral testing was also constrained by limited experience with certain IgM assays. The development, testing, and application of comprehensive broad-range viral and fungal consensus primers for use in PCR assays may be helpful. Through this project, we created a population-based bank of clinical specimens that may prove valuable in the search for newly recognized etiologic agents, the development of diagnostic tests, and the standardization of nucleic acid-based techniques for identifying previously unknown etiologic agents.

This project represents an attempt to build capacity for early detection and response to emerging infectious diseases threats in the United States and elsewhere. The usefulness of this surveillance system for UNEX was recently illustrated during an outbreak of West Nile virus encephalitis in the northeastern United States (22) and an outbreak of unexplained illness among injecting drug users in Scotland and Ireland (23); initial reports of illness from both these investigations were received through the UNEX surveillance project, and initial testing was conducted through the infrastructure developed for this project. Future surveillance for UNEX may benefit from simplified case-finding methods, improved specimen quality, and more focused syndrome-specific surveillance. Once validated, surveillance methods may be adopted by the broader public health community. Such surveillance approaches will strengthen the collaboration between clinicians, laboratorians, and public health professionals, leading to improved detection of unexplained deaths and critical illnesses, including possible bioterrorism events, and better monitoring of emerging infectious diseases.