Attributable risk of functional measures Karen Bandeen-Roche Johns Hopkins Bloomberg School of Public Health NIA Workshop on Pathways, Contributors & Correlates of Functional Impairment across Specialties August 26, 2016
Framing reminder Session topic: What do physical and cognitive assessments add to the characterization of health outcomes independent from widelyapplied measures of disease comorbidity or severity? My interpretation: How much of the risk of adverse health outcomes is attributable to poor physical / cognitive function, independent from comorbidity and etc.
What is attributable risk (attributable fraction)? Proportion of outcome risk eliminated if risk factor removed [P(Outcome)-P(Outcome if risk factor removed)]/p(outcome)] Example: Proportion of hip fracture eliminated if low walking speed removed Related: prevented fraction, mediated fraction, generalized AF Why is it important In general: Focus on intervention and public health impact Re functional measures: May motivate translation into clinical / preventative practice Why is it hard Observational versus causal Confounders, effect modifiers Causal inferences from observational data
Current state of the art - Statistical Estimating attributable risk AR= [P(Outcome)-P(Outcome risk eliminated)]/p(outcome)] A common estimate AR=[(RR-1)P(risk factor)]/[1+(rr-1)p(risk factor)] Plug in RR from Cox model, P(risk factor) from population estimate Stata software: aflogit (Greenland & Drescher, Biometrics, 1993) Elaborations Confounding adjustment: P(risk factor outcome)(rr-1)/rr, other Case-control studies Interactions Sequential and average attributable fractions (Eide & Gefeller, J Clin Epidemiol, 1995) NOT much addressed: causal associations General reference: Benichou, Statistical Methods in Medical Research, 2001
Current state of the art Gerontological Literature on functional associations with outcomes Let s grant this
Current state of the art Gerontological Attributable risk studies specifically My interpretation: % risk of geriatric outcomes attributable to poor physical / cognitive function, independent from comorbidity and etc. Methodology: Librarian-assisted literature review (thanks Lori Rosman) 1. (attributable risk* or attributable fraction* or preventable fraction* or prevented fraction*).tw. 2. exp Aged/ or exp Geriatrics/ or exp Health Services for the Aged/ 3. (ageing or latelife or laterlife or "late life" or "later life" or elder* or Geron* or Geriatr*).tw. 4. (old* adj3 (age* or adult* or population* or communit* or people* or person* or patient* or inpatient* or outpatient*)).tw. 5. ((aging or aged) adj3 (adult* or population* or communit* or people* or person* or patient* or inpatient* or outpatient*)).tw. 6. senior*.tw. 7. 2 or 3 or 4 or 5 or 6 8. 1 and 7 (1,885 hits) 23. exp Neuropsychological Tests/ or (Neuropsychological Test* or Neuropsychological Assessment* or Cognitive Test* or Cognitive Assessment*).tw. 24. 8 and 23 (12 hits) 19. (Performance or physical* or function* or gait* or walk* or grip* or grasp* or hip or knee or strength).tw. 25. exp Exercise Test/ or exp Gait/ or exp Walking/ or exp Hand Strength/ or exp Disability Evaluation/ 26. 19 or 25 27. 8 and 26 (383 hits)
Current state of the art Gerontological Attributable fraction studies specifically Results (roughly 390 articles) Roughly 210 relevant Fraction of outcome risk attributable to lifestyle 140 Fraction of functional risk attributable to disease, etc. - 34 Fraction of outcome risk attributable to measures related to function 10 Dementia, frailty, sarcopenia Fraction of outcome risk attributable to functional measures - 23
Current state of the art Gerontological Attributable fraction studies specifically Fraction of outcome risk attributable to functional measures 23 Cognitive function 4 Risk of physical disability - 3; Mortality 1 Visual function 4 Risk of: hip fracture 2; physical disability 1; driving cessation 1 Physical function 16, with risks of: Depression 4 Mortality 3 Chronic disease incidence / worsening 3 Fracture - 2 Infections 2 One each receipt of home help, falls Bottom line: Not much; plenty of room / need for further work
Example (preliminary!) Attributable fraction of SPPB for Mortality Controlling for multimorbidity (and BMI and GDS depression score) # : angina, CHF, MI, HBP, diabetes, stroke, arthritis, hip fracture, osteoporosis Also baseline age, race, education, self reported health, smoking, drinking Methods Women s Health and Aging Study II (n=436 at baseline) 7 rounds ~18 months apart Discrete (grouped) survival analysis (Prentice & Gloeckler, Biometrics, 1978) Time varying confounders versus marginal structural models Comparison of RR and AF across analyses
Example (preliminary!) Attributable fraction of SPPB for (18-mo) Mortality Results Model SPPB<5 Estimating method MSM (Causal) Direct adjustment (Naïve) Risk ratio (95% CI) AR (%) Risk ratio (95% CI) AR (%) 4.29 (2.44,7.52) 18.9 18.5 (8.4,27.5) 3.43 (1.99,5.93) 17.5 16.6 (8.1, 24.3) (Proportion of person records: 6.14% overall; 24.7% of persons dying by the next round )
What next? Identification of outcomes for which attribution of risk to functional measures is most urgently needed. Then, studies.
What next? Maximizing information regarding causal implications of functional measures for outcomes Existing cohort studies What are the key confounders? What are the key effect modifiers? Explicitly causal analyses New cohort studies? New experimental studies? Improved methods for handling repeated outcome information in AF estimation
EXTRA SLIDES
What is attributable risk (attributable fraction)? Proportion of outcome risk eliminated if risk factor removed [P(Outcome)-P(Outcome risk removed)]/p(outcome)] Example: Proportion of hip fracture eliminated if low walking speed removed Related: prevented fraction, mediated fraction, generalized AF Why is it important In general: Intervention- and public health impact Re functional measures: May motivate translation into clinical / preventative practice Why is it hard Observational versus causal Confounders, effect modifiers Causal inferences from observational data Multimorbidity Function
Example (preliminary!) Attributable fraction of SPPB for (18-mo) Mortality Results Model SPPB<9 Estimating method MSM (Causal) Direct adjustment (Naïve) Risk ratio (95% CI) AF (%) Risk ratio (95% CI) AF (%) Log-binomial 2.71 (1.67, 4.40) 40.5 2.68 (1.65, 4.32) 40.2 Log-Poisson 2.73 (1.69, 4.11) 40.7 38.4 (18.9, 53.2) 2.70 (1.67,4.36) 40.4 40.1 (20.1, 54.5) (Proportion of person records: 33.8% overall; 64.2% of persons dying by the next round ) Model SPPB<5 Estimating method MSM (Causal) Direct adjustment (Naïve) Risk ratio (95% CI) AF (%) Risk ratio (95% CI) AF (%) Log-binomial 4.31 (2.44,7.59) 19.0 3.14 (1.76, 5.61) 16.8 Log-Poisson 4.29 (2.44,7.52) 18.9 18.5 (8.4,27.5) 3.43 (1.99,5.93) 17.5 16.6 (8.1, 24.3) (Proportion of person records: 6.14% overall; 24.7% of persons dying by the next round )
What next? Identification of outcomes for which attribution of risk to functional measures is most urgently needed. Proposal re geriatric outcomes: ICHOM working group Then, studies.