Analysis of Exercise with Big Datainstruction.pstat.ucsb.edu/URCA_2016/Vitality group 2 final...

1
Analysis of Exercise with Big Data Brendon Josephson, Jake Portanova, Yuan Jiang Faculty Advisors: Ian Duncan, Janet Duncan; Statistical Advisor: Wade Herndon Dept. of Statistics and Applied Probability, University of California Santa Barbara v Data v Data from 2012-2015 (Source: Vitality) v Longitudinal data set consisting of people who consistently reported data to the Vitality program between 2013 and 2015 v 8,519 observations v Variables v Eight Health related outcome variables v Body Mass Index (BMI) v Glucose v High Density Lipoproteins (HDL) v Low Density Lipoproteins (LDL) v Total Cholesterol (TC) v Triglycerides v SBP v DBP v Verified Workout Metrics v Light Verified Workouts (LVW): 15 min at 60% max heart rate v Standard Verified Workouts (SVW): 30min at 60% max heart rate v Total Verified Workouts (TVW) = LVW + SVW v Change variables created for each of the response variables and workout categories v Models v Selection/Exploratory Analysis v ANOVA tests were used to decide which response variables are most significantly correlated with health v Collinearity: Checked the correlation between LVW and SVW to prevent inaccurate final models v Established what baseline data was relevant to the model v Final Models v Used linear models in R to predict the change in our LVW, SVW, and TVW to the change in our eight response variables v Created 72 models v (8 response variables) x (3 years) x (3 workout types) Methods v Motivation v For substantial health benefits The Office of Disease Prevention and Health Promotion (ODPHP) recommends 1 v 150 minutes of moderate-intensity physical activity v 75 minutes of vigorous-intensity physical activity v University of Kansas study on exercise and weight loss concluded 2 : v Aerobic exercise alone results in clinically significant weight loss for both men and women v There is no statistically significant difference in BMI reduction between 400 calorie workouts and 600 calorie workouts v 11.1% of US healthcare expenditure has been associated with inadequate physical activity 3 v Study Goals v Examine the effects of changes in FitBit verified data (light and standard workouts) on changes in health outcomes v Explore the effect of changing workout habits depending on previous workout habits v Analyze the effects of providing a workout incentive program (i.e. Vitality) v Increasing the amount of verified workouts causes favorable outcomes for BMI, triglycerides, and total cholesterol, but not other health measures. v Favorable health effects are attributed to an increase in SVWs, while there is little evidence that LVWs have any impact. v TVW are highly correlated with SVW but not LVW. v HDL is only significant in the two year study hinting towards the fact that a change in HDL may require long term changes to one’s physical activity habits. v US Department of Health and Human Services. 2008 physical activity guidelines for Americans. 2008. v Kokkinos, Peter. “Physical Activity, Health Benefits, and Mortality Risk.” ISRN Cardiology 2012 (2012): 718789. PMC. Web. 1 Mar. 2016. v Donnelly, Joseph E. et al. “Aerobic Exercise Alone Results in Clinically Significant Weight Loss for Men and Women: Midwest Exercise Trial- 2.” Obesity (Silver Spring, Md.) 21.3 (2013): E219–E228. PMC. Web. 1 Mar. 2016. v Summary Statistics v Mean Verified Workouts by Category v Mean and Standard Deviation of the Change Variables v Variable Relationships v Physical Activity Stratified by Gender v Model v The Regression results were most most significant in BMI and Triglycerides (results in table below) v Coefficients for Improvement in TVW throughout the study v T-Tests for Increase/Decrease in Improvement Introduction Discussion/Conclusion Works Cited [Selected] Results 2013-2014 2014-2015 2013-2015 Mean SD Mean SD Mean SD BMI 0.00 1.93 0.00 2.00 0.00 2.34 HDL 0.02 0.28 0.02 0.28 0.03 0.27 TC 0.01 0.68 0.02 0.66 0.02 0.84 TRI 0.02 0.74 0.03 0.72 0.03 0.82 LVW 0.00 4.11 0.00 3.95 0.00 2.75 SVW 0.00 5.40 0.00 4.93 0.00 3.47 TVW 0.00 6.99 0.00 6.29 0.00 4.40 Age 18-25(172) 26-50(5311) 50+(2613) Total Year M (73) F (99) M (2465) F (2946) M (1033) F (1580) (8519) LVW 2013 1.46 1.23 1.99 2.46 2.37 2.95 2.40 2014 2.93 2.49 3.38 4.00 3.56 4.34 3.81 2015 4.03 3.59 4.22 4.84 4.13 5.05 4.61 SVW 2013 4.38 2.70 5.32 4.13 6.21 4.47 4.82 2014 6.38 4.00 7.43 5.51 8.38 5.79 6.52 2015 7.76 5.10 8.52 6.30 9.53 6.58 7.45 TVW SVW LVW Year 13-14 14-15 13-15 13-14 14-15 13-15 13-14 14-15 13-15 BMI TC HDL TRI BMI and TRI Regression Output BMI Change TRI Change Coef Lower Upper Coef Lower Upper Constant 2.12 1.87 2.36 0.46 0.43 0.50 BMI (2013) -0.06 -0.07 -0.05 0.00 TRI (2013) -0.30 -0.31 -0.28 TVW (2013) 0.04 -0.03 0.10 -0.01 -0.03 0.02 TVW Improve -0.27 -0.34 -0.20 -0.06 -0.08 -0.03 Note: red valuesaresignificant 2013-2014 2014-2015 2013-2015 HDL 0.001 0.002 0.009 TC -0.030 -0.039 -0.056 Note: red valuesaresignificant

Transcript of Analysis of Exercise with Big Datainstruction.pstat.ucsb.edu/URCA_2016/Vitality group 2 final...

Page 1: Analysis of Exercise with Big Datainstruction.pstat.ucsb.edu/URCA_2016/Vitality group 2 final poster.pdf · Analysis of Exercise with Big Data Brendon Josephson, Jake Portanova ,

AnalysisofExercisewithBigDataBrendonJosephson,JakePortanova,YuanJiang

FacultyAdvisors:IanDuncan,JanetDuncan;StatisticalAdvisor:WadeHerndonDept.ofStatisticsandAppliedProbability,UniversityofCaliforniaSantaBarbara

v Datav Datafrom2012-2015(Source:Vitality)v Longitudinaldatasetconsistingofpeoplewho

consistentlyreporteddatatotheVitalityprogrambetween2013and2015

v 8,519observationsv Variables

v EightHealthrelatedoutcomevariablesv BodyMassIndex(BMI)v Glucosev HighDensityLipoproteins (HDL)v LowDensityLipoproteins (LDL)v TotalCholesterol (TC)v Triglyceridesv SBPv DBP

v VerifiedWorkoutMetricsv LightVerifiedWorkouts (LVW):15minat60%maxheart

ratev StandardVerifiedWorkouts (SVW):30minat60%max

heartratev TotalVerifiedWorkouts (TVW)=LVW+SVW

v Changevariablescreatedforeachoftheresponsevariablesandworkout categories

v Modelsv Selection/ExploratoryAnalysis

v ANOVAtestswereusedtodecidewhichresponsevariablesaremostsignificantlycorrelatedwithhealth

v Collinearity:CheckedthecorrelationbetweenLVWandSVWtopreventinaccuratefinalmodels

v Establishedwhatbaselinedatawasrelevanttothemodel

v FinalModelsv UsedlinearmodelsinRtopredictthechangeinour

LVW,SVW,andTVWtothechangeinoureightresponsevariables

v Created72modelsv (8responsevariables)x(3years)x(3workout

types)

Methods

v Motivationv ForsubstantialhealthbenefitsTheOfficeofDisease

PreventionandHealthPromotion (ODPHP)recommends1v 150minutesofmoderate-intensityphysicalactivityv 75minutesofvigorous-intensityphysicalactivity

v UniversityofKansasstudyonexerciseandweightlossconcluded2:

v Aerobicexercisealoneresultsinclinicallysignificantweightlossforbothmenandwomen

v ThereisnostatisticallysignificantdifferenceinBMIreductionbetween400calorieworkouts and600calorieworkouts

v 11.1%ofUShealthcareexpenditurehasbeenassociatedwith inadequatephysicalactivity3

v StudyGoalsv ExaminetheeffectsofchangesinFitBit verifieddata(light

andstandardworkouts) onchangesinhealthoutcomesv Exploretheeffectofchangingworkouthabitsdepending

onpreviousworkout habitsv Analyzetheeffectsofprovidingaworkout incentive

program(i.e.Vitality)

v IncreasingtheamountofverifiedworkoutscausesfavorableoutcomesforBMI,triglycerides,andtotalcholesterol,butnototherhealthmeasures.

v FavorablehealtheffectsareattributedtoanincreaseinSVWs,whilethereislittleevidencethatLVWshaveanyimpact.v TVWarehighlycorrelatedwithSVWbutnotLVW.

v HDLisonlysignificantinthetwoyearstudyhintingtowardsthefactthata changeinHDLmayrequirelongtermchangestoone’sphysicalactivityhabits.

v USDepartmentofHealthandHumanServices.2008physicalactivityguidelines forAmericans.2008.

v Kokkinos, Peter.“PhysicalActivity,HealthBenefits,andMortalityRisk.” ISRNCardiology 2012(2012):718789. PMC.Web.1Mar.2016.

v Donnelly, JosephE.etal.“AerobicExerciseAloneResultsinClinicallySignificantWeightLossforMenandWomen:MidwestExerciseTrial-2.”Obesity(SilverSpring,Md.) 21.3(2013):E219–E228. PMC.Web.1Mar.2016.

v SummaryStatisticsv MeanVerifiedWorkouts byCategory

v MeanandStandardDeviationoftheChangeVariables

v VariableRelationshipsv PhysicalActivityStratifiedbyGender

v Model

v TheRegressionresultsweremostmostsignificantinBMIandTriglycerides(resultsintablebelow)

v CoefficientsforImprovementinTVWthroughoutthestudy

v T-TestsforIncrease/DecreaseinImprovement

Introduction

Discussion/Conclusion

WorksCited[Selected]

Results

2013-2014 2014-2015 2013-2015Mean SD Mean SD Mean SD

BMI 0.00 1.93 0.00 2.00 0.00 2.34HDL 0.02 0.28 0.02 0.28 0.03 0.27TC 0.01 0.68 0.02 0.66 0.02 0.84TRI 0.02 0.74 0.03 0.72 0.03 0.82LVW 0.00 4.11 0.00 3.95 0.00 2.75SVW 0.00 5.40 0.00 4.93 0.00 3.47TVW 0.00 6.99 0.00 6.29 0.00 4.40

Age 18-25(172) 26-50(5311) 50+(2613) Total

YearM(73)

F(99)

M(2465)

F(2946)

M(1033)

F(1580) (8519)

LVW2013 1.46 1.23 1.99 2.46 2.37 2.95 2.402014 2.93 2.49 3.38 4.00 3.56 4.34 3.812015 4.03 3.59 4.22 4.84 4.13 5.05 4.61

SVW2013 4.38 2.70 5.32 4.13 6.21 4.47 4.822014 6.38 4.00 7.43 5.51 8.38 5.79 6.522015 7.76 5.10 8.52 6.30 9.53 6.58 7.45

TVW SVW LVW

Year 13-14 14-15 13-15 13-14 14-15 13-15 13-14 14-15 13-15

BMI

TC

HDL

TRI

BMIandTRIRegressionOutputBMIChange TRIChange

Coef Lower Upper Coef Lower UpperConstant 2.12 1.87 2.36 0.46 0.43 0.50BMI(2013) -0.06 -0.07 -0.05 0.00TRI(2013) -0.30 -0.31 -0.28TVW(2013) 0.04 -0.03 0.10 -0.01 -0.03 0.02

TVWImprove -0.27 -0.34 -0.20 -0.06 -0.08 -0.03

Note:redvaluesaresignificant

2013-2014 2014-2015 2013-2015

HDL 0.001 0.002 0.009

TC -0.030 -0.039 -0.056Note:redvaluesaresignificant