This chapter focuses on algorithmic methods for case detection. As we discussed in Chapter 3, the objective of case detection is to notice the existence of a single case of a disease. Case detection is a core activity of biosurveillance. Detection of an outbreak usually depends on detection of individual cases, although it is also possible to detect an outbreak from data other than case data (e.g., retail sales of thermometers or satellite imagery).
For most of the 20th century, governmental public health relied almost exclusively on the astute clinician and the clinical laboratory to detect and report cases via the notifiable disease system. Algorithms for case detection did not exist, with the possible exception of case definitions discussed in Chapter 3, which are algorithms in the sense that they are formal specifications that a clinician or epidemiologist can follow to classify an individual as a suspected or confirmed case.
Around the beginning of the 21st century, emerging diseases and the threat of bioterrorism began to stress the capabilities of the notifiable disease system. Health departments endeavored to improve it by creating web-based forms for disease reporting and electronic laboratory reporting. Their goal is to increase completeness of reporting and decrease time latencies inherent in paper-based reporting.
After the anthrax postal attacks of 2001, health departments redoubled their efforts to improve disease reporting by increasing the "astuteness" of clinicians through education and training (Gerberding et al., 2002, Hughes and Gerberding, 2002). A specialist in infectious diseases had detected and reported the first case of inhalational anthrax (Kolata, 2001); thus, the conventional wisdom was that the notifiable disease system worked. For the foreseeable future, increasing the ability of front-line physicians to diagnose rare diseases through better training was our best defense against bioterrorism. The conventional wisdom was correct insofar as case detection by clinicians is important in outbreak detection. However, it was overly sanguine about the ability of training to improve the existing capability.1
There is a limit to which additional training can improve a clinician's ability to detect and report rare diseases to a biosurveillance organization. Humans are not perfectible in this manner, as noted first by Dr. Clem McDonald, who entitled his seminal paper on physician performance Protocol-Based Computer Reminders, the Quality of Care and the Non-Perfectability of Man (McDonald, 1976b).2
His research, conducted in the early 1970s at the Regenstrief Institute in Indiana, demonstrated that even for common conditions, such as diabetes and hypertension, and for common preventive measures, such as immunizations, physicians often failed to deliver required services. However, when reminded by a computer system that monitored electronic patient data about the need to vaccinate a specific patient or order a needed test, physicians complied with standards of care at twice the rate as when not reminded. When the system stopped reminding the physicians, however, their compliance rates quickly returned to baseline; thus, any "education" or "training" that the system provided had no lasting effect on compliance.
McDonald's research spawned a new line of system-level thinking in medicine that continues to this day (Kohn et al., 2000, Leavitt, 2001,Yasnoff et al., 2004). McDonald concluded this influential paper with the observation that, although man is not perfectible, systems of care are.
1 Studies of continued medical education programs also suggest that the yield is low and the cost is high (Haynes et al., 1984, Leist and Kristofco, 1990, Williamson et al., 1989, McDonald, 1983). The one study of efficacy of training (of a web-based educational program) on physicians' knowledge about diagnosis and management diseases caused by known weaponized biological agent showed no retention of information (chung et al., 2004).
2 We do not know whether the always playful Dr. McDonald and the sophisticated editors of the New England Journal of Medicine misspelled the word perfectibility intentionally. Also reinforcing the main conclusion of his research, it took a computer's spell checker to bring this error to our attention.
Handbook of Biosurveillance ISBN 0-12-369378-0
Elsevier Inc. All rights reserved.
2. PERFECTING CASE DETECTION_
The best case detection system imaginable would be one in which every individual in a community is examined every morning by the best diagnostician in the world. This diagnostician would have all the time in the world to interview and examine each person. Since she would be examining everyone every day, she would notice patterns (clusters) of early illness in a community. Her awareness of patterns would appropriately bias her diagnostic thinking (and treatment) of individual patients. Physicians are taught (and reminded incessantly) "when you hear hoof beats, don't think of zebras." This adage is an informal statement that when the evidence available about a particular patient supports equally a diagnosis of either influenza or SARS (e.g., the patient has constitutional symptoms and no history of exposure to SARS), they should conclude that the diagnosis of influenza is far more likely than SARS. This diagnostician would also never fail to report immediately each fever, early syndromic presentation, or reportable disease to governmental public health.
This ideal scenario recognizes the importance of the knowledge, judgment, and skills that clinicians bring to bear on the diagnosis of disease. The importance of expert knowledge and judgment underlies the opinion, expressed by experts in public health after the anthrax letters in 2001, that there is no substitute for an astute clinician. This ideal scenario also recognizes that knowledge of the prevalence of disease in a population influences the diagnostic work up and management of individual patients.
Of course, such a system is not feasible due to the impossibility of having every person seen by the same diagnostician every day or the alternative of cloning the best diagnostician and ensuring that the clones could instantly share information about individuals they were seeing. Nevertheless, it represents a benchmark against which we can compare other schemes that are perhaps superior to current approaches.
In the next section, we discuss diagnostic expert systems, which are computer programs (algorithms) that embody the diagnostic knowledge and diagnostic skills of expert clinicians. Diagnostic expert systems are far less expensive than clinicians, never tire, and we can clone them at will. They make it reasonable to imagine a biosurveillance system in which a highly competent diagnostician examines thousands of individuals in a community every day with consistent diagnostic quality—reporting fevers, syndromes, and reportable diseases to a health department without fail and without delay. They make it possible to imagine a biosurveillance system in which the health department analyzes highly improved case data and communicates up-to-the-minute information about patterns of illness in the community back to the diagnosticians.
3. DIAGNOSTIC EXPERT SYSTEMS_
Diagnostic expert systems are computer algorithms that automate the cognitive process of medical (or veterinary) diagnosis.
Many readers will be surprised to learn that researchers in the fields of artificial intelligence and medical informatics have been developing and fielding such systems since the 1960s. The first fielded system, developed by Dr. Homer Warner in Salt Lake City, provided diagnostic assistance for children with congenital heart disease (Warner et al., 1961, Warner et al., 1964). Congenital heart diseases are severe birth defects of the valves and structure of the heart. In the 1960s, the exact nature of heart malformations was very difficult to diagnose without invasive and risky angiographic procedures. Dr. F. Timothy de Dombal developed a similar system for the differential diagnosis of the acute abdomen, another high-stakes diagnostic problem. A surgeon must differentiate between conditions that require emergency surgery, such as appendicitis, and conditions, such as pancreatitis, for which surgery is relatively contraindicated (de Dombal et al., 1972, de Dombal et al., 1974, Wilson et al., 1975, de Dombal, 1975,Wilson et al., 1977, de Dombal, 1984,Adams et al., 1986, McAdam et al., 1990, de Dombal, 1990, de Dombal et al., 1993, American College of Emergency Physicians, 1994).
3.1. How Diagnostic Expert Systems Work: Data Collection
We can perhaps best explain how a diagnostic expert system works by analogy to the process that a physician uses to diagnose a patient. Like a physician, a diagnostic expert system begins by collecting information about the patient's illness— symptoms, observations from physical examination, results from laboratory tests, risk factors for disease (e.g., travel to a foreign country) and pre-existing medical conditions (e.g., diabetes). We refer to this diagnostic information collectively as the findings. Of course, the computer usually does not interview the patient and (at present) never examines the patient. Rather, a physician interviews and examines the patient after which she or an assistant enters the findings into the program (e.g., Warner and Bouhaddou, 1994, London, 1998, Buchanan and Shortliffe, 1984, Miller et al., 1986, Shwe et al., 1991, Heckerman et al., 1992). Increasingly, diagnostic expert systems acquire findings automatically from clinical information systems (Aronsky et al., 2001, Burnside et al., 2004, McDonald et al., 1991). There are also examples of diagnostic expert systems that interview patients directly to obtain their medical histories (Pynsent and Fairbank, 1989, Wald et al., 1995).
3.2. How Diagnostic Expert Systems Work: Knowledge Representation
Like the physician, the diagnostic expert system is a storehouse of medical knowledge. A diagnostic expert system stores its medical knowledge in tables of diseases and their findings. There are typically a table of prevalences for each disease (e.g.,Table 13.1) and tables with every finding of every disease that the system knows about (e.g., Table 13.2). The latter tables usually represent the strength of association between diseases and findings as conditional probabilities.
table 13.1 Prior Probabilities and Prior Odds of FMD and MCD
Was this article helpful?
All you need is a proper diet of fresh fruits and vegetables and get plenty of exercise and you'll be fine. Ever heard those words from your doctor? If that's all heshe recommends then you're missing out an important ingredient for health that he's not telling you. Fact is that you can adhere to the strictest diet, watch everything you eat and get the exercise of amarathon runner and still come down with diabetic complications. Diet, exercise and standard drug treatments simply aren't enough to help keep your diabetes under control.