Paper
Studying Symptoms: Sampling and Measurement Issues
Published May 1, 2001 · K. Kroenke
Annals of Internal Medicine
184
Citations
6
Influential Citations
Abstract
Physical symptoms account for more than half of outpatient encounters in the United States, or almost 400 million clinic visits each year (1). An exact medical diagnosis that explains the symptom is often not established; at least one third of symptoms lack an adequate physical explanation (2-4) and are referred to by various labels, including functional, idiopathic, atypical, somatoform, or unexplained. Even without a disease moniker, physical symptoms produce impairment in patient functioning and quality of life similar to that seen with many better-defined medical and psychiatric diseases (4-9). Physical symptoms generate considerable health care expenditures in terms of clinic visits, laboratory testing, medications and other therapies, and subspecialty referrals (10). Yet therapy for symptoms is often disappointing, causing patient and provider dissatisfaction (11, 12). The costs and prevalence of, the impairment caused by, and the diagnostic and therapeutic uncertainty surrounding common symptoms make these symptoms a priority for primary care research (13). In this paper, I examine eight sampling and measurement issues central to the study of symptoms. Rates Symptoms are ubiquitous, and chief complaints are but a fraction of them. In addition to being the leading reason for outpatient visits, symptoms are an associated or undeclared problem in many other clinic patients. Furthermore, most persons who have symptoms never report them to a physician, resulting in a large reservoir of symptoms in the community (14-16). For example, fatigue was found to be the major reason for physician visits in 6.7% of patients attending a primary care clinic, but it was a subsidiary symptom in 13.6% (17). Moreover, when all patients attending a primary care clinic were surveyed, more than 20% endorsed fatigue as a major problem (18). Clinic surveys show that the endorsement rate for many common symptoms exceeds 10% to 20%, that fewer than one in five clinic attendees are symptom free, and that the typical outpatient identifies two to four symptoms as bothersome (4, 12, 19). This considerable background prevalence complicates the interpretation of symptoms. When attributing a common symptom (such as headache) to a common disorder (such as hypertension), a coincidental association may be as likely as a causal one. Similarly, before labeling symptoms a side effect of a medication, one must recognize that many patients report similar symptoms while receiving either a placebo or no medication (20). The high base rates of symptoms mandate explicit criteria for the selection of case-patients and controls in symptoms research. For example, when conducting a study of chronic fatigue, the investigator must decide whether to include as case-patients (and exclude as controls) only patients with a chief complaint of fatigue or to extend eligibility to those with any symptoms of fatigue, whether reported spontaneously or elicited on interview. Population Selection bias can be a serious concern in symptoms research. Symptomatic patients referred to subspecialty clinics represent selected cases and may differ from patients seen in the general medical setting. Atypical, persistent, or complicated symptoms may survive the referral filter, whereas self-limited, mild, or classic cases are managed predominantly by the primary care physician. To identify the common causes of symptoms and a reasonable initial evaluation, we need more studies of all-comers. Selection bias affects both standard subspecialty clinics and research clinics set up to study specific symptoms, such as fatigue clinics, headache clinics, and dizziness clinics. When patients are not drawn from a primary care setting, the investigators should describe how they were referred and what proportion of the population of interest they represent. Exclusion criteria should be limited and readily defensible, and nonparticipants should be characterized so that research findings can be interpreted in the context of all patients who might present with a given symptom. An exception to this inclusive approach would be a study that targets not all patients with a given symptom but only those with chronic or disabling symptoms. An even larger group of symptomatic persons resides outside of the clinics, since symptoms in the community are often considered normal phenomena; resolve spontaneously; or are managed with self-treatment, lay consultation, or alternative medicine (3, 14, 15). Although overmedicalization of these symptoms should be avoided (10, 21), studies comparing symptomatic persons who do use health care with symptomatic persons who do not may identify determinants that escalate symptoms into illness (22). Detection Three methods to detect symptoms in clinical research are 1) chart review [discovery of what the provider has documented in the patient's record]; 2) elicitation by a survey, questionnaire, symptom diary, or a provider's review of systems; and 3) spontaneous reporting by the patientthat is, a volunteered symptom, often labeled a chief complaint or presenting complaint. Each method may detect symptoms of a different nature and threshold. Reliance on what is noted in the patient's chart probably underestimates symptom burden: A comparison of chart review studies (2, 23) with clinic surveys (4, 12, 19) suggests that symptoms often go unrecorded. Patient surveys or structured interviews can capture symptoms that physicians fail to recognize or record, but they may exaggerate true morbidity because of overendorsement bias, a tendency for patients to generously claim the symptoms asked about on a checklist. In four studies using checklists of 15 to 20 symptoms, clinic patients endorsed a median of three to four symptoms as bothersome (4, 12, 19, 24). Clinically relevant symptoms become nested in a larger number of affirmed symptoms, past and present, troublesome as well as trivial. Volunteered symptoms, those presenting as chief complaints, are probably the category most relevant to clinical practice. However, the patient's personality characteristics as well as the provider's interviewing skills, clinical interests, and available visit time may alter the threshold for the spontaneous report of physical complaints. In addition, other factors compete for space in the time-limited agenda of an ambulatory encounter; these factors include chronic medical conditions, health maintenance, prescription refills, and the ordering of tests or procedures. Although clinicians prefer to concentrate on a single reason for the encounter, patients typically have several concerns other than their chief complaint. In short, volunteered, elicited, and documented symptoms each provide a different perspective of illness behavior and clinician response. Temporal Factors Operational criteria with respect to time course are particularly critical in symptoms research. Recency of onset is one consideration in determining study eligibility. In some studies, patients with chronic symptoms might be excluded so that the study focuses on incident cases. One way to do this is to include only patients with current symptoms; another is to include patients who have had symptoms within a certain period, such as the past 7 days or the past month. In some studies, it may be desirable to exclude transient or self-limited symptoms by requiring that symptoms have persisted for a minimum time period, such as 2 weeks. Episodicitythe fact that symptoms are not present continually but rather come and go in discrete blocks of time or episodes, with symptom-free periods between episodesis another temporal factor. In one study, only 4% of dizzy patients reported constant symptoms, fewer than half had daily spells, and more than 20% had dizziness less than once a week (25). Indeed, patients with Mnire disease may go months or even years without an attack. Migraine headache, low back pain, the irritable bowel syndrome, and many other conditions can be intermittent. Sporadic symptoms should be quantified by frequency and duration of episodes. Episodicity must be accounted for in evaluating outcomes in a clinical trial; a therapeutic response might include a reduction in the number, duration, or severity of episodes. Episodicity must also be considered when an inception cohort is being assembled for a study of natural history. One might include only patients having their first episode or, alternatively, might stratify the analysis according to whether symptoms are new or recurrent. A third temporal issue is the duration of symptomatic episodes. Unlike many medical disorders that are present continually, symptoms may come and go, lasting seconds, minutes, hours, or days. The frequency and duration of spells can help assess the cause and severity of symptoms, as well as monitor treatment response. What is the optimal control group in symptoms research? Can one include persons who have had the symptom in the past but are currently asymptomatic, or should eligibility be restricted to those who have never had the symptom? In one clinic survey, 21% of patients currently had dizzy spells, 20% had had dizziness in the past, and 59% said that dizziness had never been a problem (6). Although the last group may be the purest control group, some symptoms, such as headache and fatigue, are so universal that investigators may decide to include as controls those persons who have been symptom-free for a minimum period. Severity When is a symptom an illness? Temporal features, such as frequency and duration of episodes, make up one threshold. A second threshold is severity. The measurement of symptom severity relies almost entirely on patient self-report. Even when objective measures are available, they may not accurately reflect symptom severity. In chronic obstructive pulmonary disease, dyspnea as experienced by the patient may correlate poorly with airway obstruction as measured by spirometry. Similarly, back pain and imaging abnormaliti
Highly CitedSymptom research requires careful selection of case-patients and controls due to high prevalence and potential population selection biases.
Full text analysis coming soon...