Online Supplement
Physical Function as a Prognostic Biomarker among Cancer Survivors
SUPPLEMENTARY METHODS
Objective Measures of Physical Function
All physical function measures were completed at various times of the day inmobile examination centers (93% of participants) or in the participant’s home (7% of participants) using standardized requirement, procedures, and scriptsat both locations [Ostchega et al, 2000]. Individuals tested in the mobile examination center were significantly younger, less likely to be women, and less likely to report needing help with personal care and routine daily needs when compared to those tested at home (all P<0.001) [Ostchegaet al, 2000]. Five percent of individuals were invited back 2-3 weeks later for reexamination. All timed physical function measures had reliable intra-class correlation coefficients upon reexamination (ICC > 0.5). There were no reports of injury or any other adverse effects for study participants during the physical performance measures [Ostchegaet al, 2000].
The Short Physical Performance Battery
The three tests in the short physical performance battery (SPPB) include: 1) time required to walk 2.4-meters at a usual pace; 2) time required to rise five times from a chair with the arms folded in front of the chest; and 3) ability to complete three standing measures that include balancing side-by-side, semi-tandem, and full tandem, each for 10-seconds. Modifications to the SPPB used in NAHNES III did not include the side-by-side and semi-tandem balance measures (assessing only full-tandem balance), and measured gait at a pace twice that of the usual walking speed (versus usual speed), referred to as fast walk speed, as described in detail below.
Fast Walk Speed
Fast walk speed was assessed using the fastesttime of two 2.4-meter walks. Participants were instructed to complete the walk at a pace twice that of their usual walking speed [Ostchega et al, 2000]. Time required to complete the 2.4-meter course was recorded to the nearest tenth of a second using a stopwatch.
Mortality Outcome
Participants were linked to the National Death Index (NDI) database using a probabilistic matching algorithm that included 12 identifiers including Social Security Number, sex, date of birth, race, state of residence and birth, and marital status [Rogot et al, 1986]. The National Center for Health Statistics found that 96.1% of deceased participants and 99.4% of living participants were correctly classified using the probabilistic matching algorithm [National Center for Health Statistics, 2006a]. Study participants who were not matched with a death certificate were categorized as being alive through the end of the follow-up period. The National Center for Health Statistics removed select subject characteristics in the file to prevent the re-identification of study participants. The publically released survival data are nearly identical to the restricted-use NHANES III linked mortality file [Wilper et al, 2009].
Covariates
Demographic information including date of birth and sex were self-reported using a standardized questionnaire [National Center for Health Statistics, 2006c]. Clinical information including type of cancer, date of cancer diagnosis, smoking history, alcohol consumption, hospitalizations in the prior year, self-reported health status (using the Short Form 36), and frequency of physical activity were assessed using standardized questionnaires [National Center for Health Statistics, 2006c]. The presence of comorbid health conditions were determined by asking participants if a doctor had ever told them that they had any of the following: hypertension, diabetes, hyperlipidemia, asthma, arthritis, myocardial infarction, stroke, or congestive heart failure.
Height in meters and body mass in kilograms were measured by study technicians. Body mass index was calculated as body mass divided by the square of height (kg/m2). Bouts of walking in the past week were self-reported and included any bout of walking that was estimated to be ≥1 milein duration, and of moderate or vigorous intensity. The healthy eating index (HEI) was calculated from 24-hour food recalls to form a score than ranges from 0 to 100 to quantify aspects of a healthy diet [Kennedy et al, 1995]. Hemoglobin was quantified using a Coulter S-Plus Jr electronic counter (Coulter Electronics, Hialeah, FL) with a coefficient of variation <3.0%. Albumin was quantified using a Hitachi 737 multichannel analyzer (Boehringer Mannheim Diagnostics, Indianapolis, IN, USA) with a coefficient of variation of 2.8%. C-reactive protein was quantified using latex-enhanced nephelometry immunoassay (Behring Diagnostics, Somerville, NJ) with a coefficient of variation of 6.3%. Detailed blood collection procedures and laboratory assay methods used in NHANES III are described elsewhere [Lacher et al, 2005, National Center for Health Statistics, 2006b].
Statistical Analysis
To test if the prognostic importance of physical function varied as a function of type of cancer, we calculated statistical interactions with the physical function measure and type of cancer (i.e., Pinteraction). The assumption of proportional hazards for all analyses was confirmed using log-log plots.
To quantify the discriminative capacity of the SPPB and fast walk speed to predict five- and ten-year mortality, we calculated the area under the receiver operating characteristic (ROC) curve, known as the C-statistic, using logistic regression models that adjusted for age and sex. The C-statistic ranges from 0.5 to 1.0, with values of 0.6, 0.7, 0.8, and 0.9 representing poor, fair, good, and excellent discrimination, respectively.
Sample weights were incorporated into all statistical analyses to account for nonresponse bias, multistage sampling probabilities, and the subpopulation of participants that completed the physical performance evaluation [Korn and Graubard, 1991].
SUPPLEMENTARY RESULTS
Objective Physical Function and All-Cause Mortality, Interaction by Cancer Type
Type of cancer did not modify the prognostic importance of the SPPB as a continuous or categorical variable (Pinteraction=0.410 and Pinteraction=0.766, respectively). Type of cancer did not modify the prognostic importance of fast walk speed as a continuous or categorical variable (Pinteraction=0.825 and Pinteraction=0.863, respectively).
Discriminative Capacity of the SPPB and Fast Walk Speed to Predict Five- and Ten-Year Mortality
During the first five-years of follow-up, 136 (33%) people died. The SPPB and fast walk speed demonstrated fair discrimination predicting five-year mortality. The C-statistics for the SPPB and fast walk speed were 0.75 (95% CI: 0.70–0.80) and 0.74 (95% CI: 0.70–0.79), respectively, and did not statistically differ (P=0.64). During the first ten-years of follow-up, 235 (57%) people died. The SPPB and fast walk speed demonstrated good discrimination predicting ten-year mortality. The C-statistics for the SPPB and fast walk speed were 0.82 (95% CI: 0.78–0.86) and 0.83 (95% CI: 0.79–0.87), respectively, and did not statistically differ (P=0.50).
ONLINE SUPPLEMENT REFERENCES
Kennedy ET, Ohls J, Carlson S, Fleming K (1995) The healthy eating index: design and applications. J Am Diet Assoc 95: 1103-1108
Korn EL and Graubard BI (1991) Epidemiologic studies utilizing surveys: accounting for the sampling design. Am J Public Health 81: 1166-1173
Lacher DA, Hughes JP, Carroll MD (2005) Estimate of biological variation of laboratory analytes based on the third national health and nutrition examination survey. Clin Chem 51: 450-452
National Center for Health Statistics (2006a) The Third National Nutrition and Health Survey Linked Mortality File: Matching Methodology (
National Center for Health Statistics (2006b) Laboratory Procedures Used for the Third National Health and Nutrition Exam Survey (NHANES III), 1988-1994 (
National Center for Health Statistics (2006c) NHANES III Questionnaires, Datasets and Related Documentation (
Ostchega Y, Harris TB, Hirsch R, Parsons VL, Kington R, Katzoff M (2000) Reliability and prevalence of physical performance examination assessing mobility and balance in older persons in the US: data from the Third National Health and Nutrition Examination Survey. J Am Geriatr Soc 48: 1136-1141
Rogot E, Sorlie P, Johnson NJ (1986) Probabilistic methods in matching census samples to the National Death Index. J Chronic Dis 39: 719-734
Wilper AP, Woolhandler S, Lasser KE, McCormick D, Bor DH, Himmelstein DU (2009) Health insurance and mortality in US adults. Am J Public Health 99: 2289-2295