Over the past 50 years, cancer research has emphasized individual behaviors (e.g., early detection screening, smoking, alcohol use, diet and nutrition, and physical activity) as important foci for prevention (Hataway & Bragg, 1984). Self-report instruments are among the primary methods of assessing cancer-related variables, including early detection screening and behavioral risk factors (primary prevention), as well as psychosocial risk factors (secondary and tertiary prevention). This section discusses self-reported cancer screening and considerations for the use of self-reported behavior and psychosocial risk factors, and concludes with suggestions about how to most effectively use and interpret self-report data.
Self-reported cancer screening. Cancer screening is a commonly reported clinical assessment designed to facilitate early detection, and regular screening is important for reducing morbidity and mortality across an array of cancer types. Although more objective alternatives exist for determining screening practices in the population (e.g., health insurance or medical records), self-report of screening is nonetheless the measure of choice in the majority of studies. For example, about two-thirds of research examining Pap smear testing utilized self-report methods (Newell, Girgis, Sanson-Fisher, Savolainen, & Hons, 1999).
With the growing emphasis on preventive care, a body of research has tested strategies aimed at encouraging patients to be screened in accordance with recommended guidelines (e.g., Masi, Blackman, & Peek, 2007). Although increased screening is not universally recommended, efforts to promote screening are common – including, but not limited to, screening for breast, cervical, ovarian, colorectal, and prostate cancers. Self-report validity studies often report kappa (i.e., percent agreement), sensitivity (i.e., true positives—the proportion of participants screened according to objective markers who self-reported the screening), and specificity (i.e., true negatives—the proportion of participants not screened according to objective markers who self-reported no-screening). Concordance rates for medical records and self-reported data indicate that the prevalence of cancer screening is overestimated by self-report (Gordon, Hiatt, & Lampert, 1993; Hiatt et al., 1995), whereas the time since the most recent test is underestimated (Gordon et al., 1993). Research in other areas has found that people can often accurately report the day of week a given event occurred, yet they tend to report a more recent date than actually was the case (Cohen & Java, 1995; Thompson, Skowronski, & Lee, 1988). This may occur because greater clarity of a memory provokes feelings of recency (Bradburn et al., 1987). Along these lines, people tend to anchor their reports to reasonable timeframes and/or round the values off to the number of weeks or months (Huttenlocher et al., 1990). Each of these lines of evidence suggests that report of specific dates or the duration of time passed since an event may be inaccurate. The sources of these biases are unclear, but they likely reflect individuals’ reliance on schemas when answering temporally-based questions.
Although findings are mixed, the utility of self-report data for some specific types of screening is promising. Some research has found high sensitivity and agreement for breast and cervical cancer screening (Caplan et al., 2003), although other research has not (Bowman, Redman, Dickinson, Gibberd, & Sanson-Fisher, 1991; Bowman, Sanson-Fisher, & Redman, 1997). Lykins, Pavlik, and Andrykowski (2007) concluded that the validity of self-report for determining routine ovarian cancer screening (i.e., transvaginal sonography; TVS) was very high when compared to medical records. Indeed, their evidence suggests that TVS self-reports are more accurate than breast, cervical, and colorectal cancer screening reports. Despite these encouraging findings, however, accuracy of self-reported mammography and Pap smear testing for clinical decision-making may be lower among low socioeconomic, underinsured, and/or minority groups (McGovern, Lurie, Margolis, & Slater, 1998; McPhee et al., 2002; Vacek, Mickey, & Worden, 1997). Fewer studies have examined the use of self-report for colorectal or prostate cancer screening, but existing evidence suggests high (Bleiker et al., 2005), or fair to moderate concordance with medical records (Hall et al., 2004; Jordan, Price, King, Masyk, & Bedell, 1999).
Overall, research suggests that patient self-reports of cancer screening are reasonably valid. However, the precision of estimates of timing are considerably less reliable (e.g., specific dates or the time since the most recent screening). It is important to note, however, that patient reports are not always intentionally biased. For example, several types of screening can be conducted as part of a full examination, leaving patients unaware that a particular test was conducted (Hall et al., 2004).
Self-reports of cancer risk behaviors. Cancer risk is elevated by the presence of both uncontrollable and controllable risk factors. Self-reports are used to estimate the prevalence of risk factors in the population and the efficacy of interventions seeking to reduce them. Risk factors that are beyond one’s control generally include family history/genetics, race/ethnicity, prior history of cancer, and age. However, there are a number of controllable risk factors that are related to behavior, such as smoking, heavy alcohol use, poor diet/nutrition, physical inactivity, ultraviolet light exposure, and risky sexual behavior. A complete review of the nature of self-reports of these factors is beyond the scope of this discussion; rather, our aim is to consider several key variables related to cancer risk, and to briefly describe evidence for the validity of their self-report. Newell and colleagues (1999) conducted a comprehensive review of the accuracy of self-reported health behaviors and cancer-related risk factors. They found that, in general, self-reports consistently underestimated risk factor prevalence and percentages of ‘at-risk’ individuals.
Tobacco smoking is the leading cause of preventable cancer morbidity and mortality, making smoking status into one of the most commonly reported cancer-related risk factors. Earlier research revealed a tendency for smokers to underreport (e.g., Haley & Hoffman, 1985) or deny their smoking completely (e.g., Luepker, Pallonen, Murray, & Pirie, 1989; Murray, O’Connell, Schmidy, & Perry, 1987). More recently, however, evidence suggests that smokers are willing to self-disclose this behavior. Patrick and colleagues (1994) found that the overall validity of smoking self-reports is high, with sensitivity and specificity estimates of close to 90%. Self-reports of cigarette smoking also appear to be valid among adolescents across racial/ethnic groups (Kentala, Utriainen, Pahkala, & Mattila, 2004; Wills & Cleary, 1997). Some recent work, however, suggests that the accuracy of self-reports may be declining, perhaps in conjunction with the underreporting of illicit drug use (Fendrich, Mackesy-Amiti, Johnson, Hubbell, & Wislar, 2005).
Clinical trials using self-reported smoking abstinence as an outcome should follow recommended definitions (c.f. Hughes et al., 2003). Whenever possible, researchers should biochemically validate smoking status using gold standard markers, such as carbon monoxide (CO), salivary cotinine, or thiocyanate. Researchers should also consider the possible role of social desirability in smoking status reports. In particular, certain types of smokers may have a higher likelihood of underreporting, including those with chronic diseases (Fisher, Taylor, Shelton, & Debanne, 2007), pregnant women (Russell, Crawford, & Woodby, 2004), and hospital inpatients (Schofield & Hill, 1999).
Heavy alcohol use is associated with elevated cancer risk (Cargiulo, 2007). The gold standard for biochemical verification of short-term alcohol use is a breathalyzer test; yet this test is infrequently used in research (see Newell et al., 1999). Grønbaek and Heitmann (1996) found overall agreement between self-reports on an alcohol use frequency questionnaire and dietary interviews. However, the number of studies that carefully examine the validity of self-report of alcohol use is not yet adequate to form firm conclusions. Even in the absence of biochemical verification, the "bogus pipeline" (a procedure that induces the [false] belief that drinking will be biochemically verified; e.g., Botvin, Botvin, Renick, Filazzola, & Allegrante, 1984; Campanelli, Dielman, & Shope, 1987; Jones & Sigall, 1971) can increase the accuracy of self-reports of alcohol consumption in clinical or research settings.
Physical activity levels are also commonly estimated in cancer prevention studies. Reporting levels of physical activity is challenging for several reasons. Respondents are asked to recall many separate events over a period of time, some of which may not be particularly salient or memorable (e.g., walking). Recall is also more difficult when respondents report both duration and intensity of each activity. In addition, the categories of physical activities defined in surveys (e.g., the Seven Day Activity Recall questionnaire; Sallis, Buono, Roby, Micale & Nelson, 1993) require respondents to make judgments about what constitutes moderate, hard, and very hard activities. Sallis and colleagues (1993) found lower reliability for repeated reports from longer intervals (4–6 days between interviews) versus shorter intervals, suggesting rapid decay of subjects’ ability to remember specific physical activities.
There are a number of approaches to objectively assess the validity of self reports of physical activity, including: mechanical or electronic monitors (including accelerometers and heart rate monitors), energy expenditure (including doubly labeled water and calorimeter), measures of fitness, and direct observations (Kohl, Fulton & Caspersen, 2000). These types of measures, however, correlate rather weakly with self-report (Sallis & Saelens, 2000), although this weak relationship may be due in part to devices such as accelerometers missing some light and moderate activities (Richardson, Leon, Jacobs, Ainsworth & Serfass, 1995). Similarly, self-reported data on physically demanding activities are well-validated using heart-rate monitors, but lower correlations between self-reported physical activity and heart rate have been found for less intensive activities (Janz, Golden, Hansen & Mahoney, 1992). Overall, reliance on the accuracy of physical activity reports may be acceptable for specific activity frequencies (e.g., frequency of tennis) or gross activity (e.g., sedentary versus not), but far less reliable for any specific intensity measure (e.g., total time spent exercising, percent of activity at VO2max, etc).
There is a long history of research relating cancers to dietary factors (e.g., Doll & Peto, 1981), but evidence for a causative role of diet in most cancers is limited. That is, epidemiological studies have identified relationships between dietary practice and cancer development, but prospective or interventional studies have often not provided strong corroborative support. High-fat diets appear associated with increased risks of breast, colon, prostate, and endometrial cancers (USDHHS, 1988). Diets high in salt and red/processed meats have been linked with stomach and colorectal cancer, respectively (Key, Allen, Spencer, & Travis, 2002). In contrast, high fiber diets are related to reduced risk of colon cancer (Trock Lanza, & Greenwald, 1990). Unfortunately, there is no established gold standard for the measurement of diet or nutrition, leaving self-report methods open to include daily or weekly diaries, clinical interviews, and portion-size estimates. The latter approach is less frequently used and appears to offer little incremental validity to diet-related risk assessments (Paiva, Amaral, & Barros, 2004). Much research has employed food frequency questionnaires (FFQs), which ask respondents to indicate their "usual" food intake over a weekly, monthly, or yearly reference period (e.g., Zulkifli & Yu, 1992). Cavadini and colleagues (1999) found good agreement between an FFQ and "diet records" collected by experience sampling methods (see later section), although the level of agreement varies widely by type of food. However, diet records are not consistently in good agreement with an objective measure such as doubly labeled water (a method in which participants drink treated water that allows for measuring metabolic rate over days or weeks; Livingstone et al., 1992). Indeed, studies have found substantial underreporting of food intake in diet records among obese participants, female endurance athletes, and adolescents (Schoeller, 1995). These studies suggest that diet records themselves should not be used as independent methods of validation of FFQs. Other studies have found 24-hour recalls obtained via interview with a dietician provide more accurate estimates (e.g., Field et al., 1998; Frank et al., 1992; Rockett et al., 1997).
Obesity is an established risk factor for some cancers (colon, breast, endometrial, and possibly other cancers; e.g., Trentham-Dietz, Nichols, Hampton, & Newcomb, 2006; Verreault, Brisson, Deschenes, & Naud, 1989), making it important to accurately assess obesity and body fat distribution. Body mass index (BMI), or the ratio of weight to height, is often reported in epidemiological research. Although there is a tendency for overestimation of height and underestimation of weight, there is evidence for the validity of BMI self-reports (e.g., Palta, Prineas, Berman, & Hannan, 1982). Body fat distribution is measured in a variety of ways, but one easily implemented measure is waist-to-hip circumference, also referred to as waist-to-hip ratio (WHR). Weaver and colleagues (1996) found that WHR can be measured accurately by self-report when respondents are provided with all the materials needed to conduct the measurements and are given explicit instructions and training. Even under these more optimal conditions, however, women with larger WHR are likely to underestimate their measurements when compared with clinic-based assessments (Weaver et al., 1996).