Real world applications of Research Methodology & Statistics (Yes, they can be useful) Quantitative Research in the Social & Health Sciences Presentation by Ryan Pohlig, University of Scranton ’05 

Research Methods & Statistics n  Present some of the research I have personally been involved in during grad school n  Going to start with the Research Question (RQ) and background, then cover the design, data, and analysis used ¡  RQ should always drive the design & analysis n  Going to cover a wide range of topics, feel free to interrupt to ask questions 

Nursing – Pancreatic Cancer (PC) n  RQ: Want to find if there is a relationship between Quality of Life (QOL) and Cytokine levels in PC patients n  Imbalance between pro-and anti-inflammatory cytokines can cause chronic immune activation and inflammation n  Relationship between immune fluctuations and symptom burden not studied in PC patients n  Knowing these relationships, health care workers can plan for management of pain and other symptoms 

Background n  Pancreatic Cancer (PC), 4th most common cancer and it’s deadly ¡  1 year survival rate is 24% ¡  5 year survival rate is 5% ¡  Resectable 5 year rate is 17% n  This cancer spreads to distant sites early n  Unfortunately, most patients present at advanced stages at their diagnosis n  Chemo and radiation therapy are standard treatment for both localized & metastatic 

Cytokines n  Cytokines are circulating proteins produced by immune system ¡  Pro-inflammatory cytokines: IL-8 and TNF-α ¡  Anti-inflammatory cytokines: IL-10 ¡  Both pro and anti: IL-6 n  Interleukin (IL) and Tumor Necrosis Factor-alpha n  Previous research has shown elevated levels of cytokines in cancer patients relative to healthy people n  Cytokine levels are related to pain, weight loss, neuropathic inflammation, and fatigue 

Quality of Life n  Unfortunately, treatment and disease affect QOL as both create significant toxicities ¡  Which can have a profound effect on patient’s symptom burden and immune function n  Stages of Cancer (I-IV) are also directly related to QOL n  (FHSI-8) The Functional Assessment of Cancer Therapy: Hepatobiliary Symptom Index-8 ¡  A measure of Quality of Life ¡  Likert scale of pain ratings, fatigue, and symptom presence ¡  Higher scores indicate more pain & discomfort 

Study Design & Analysis n  Design- observational study ¡  Examine relationship between FHSI-8 and Cytokine levels ¡  Need to worry about cancer stage as confounder n  Analysis- Regression ¡  Control for stage of cancer, sex, & age n  Predict FHSI with controls ¡  Then test if adding Cytokine levels leads to significantly better fit (prediction) n  First Block- Covariates n  Second Block- Cytokines 

Sample n  n =36, newly diagnosed PC Patients n  Blood Samples drawn and filled out the FHSI-8 n  Sample: 19 males and 17 females n  Stage Age I 4 40-59 8 II 9 60-69 10 III 7 70-79 9 IV 16 80-89 9 

Results n  Significant improvement seen by adding Cytokines ¡  R2 = .142 to .495, p = .013 ¡  R2 can be thought of as % of variance that can be accounted for by model n  Slope Interpretations ¡  The more IL-6 the lower QOL ¡  The less IL-8 the lower QOL ¡  IL-10 not related to QOL ¡  The more TNFα the lower QOL Variable b p-value Age .237 .024 Sex 2.548 .205 Stage .359 .699 IL6 4.111 .039 IL8 -7.907 .004 IL10 -1.755 .327 TNFα 10.619 .002 

Cytokine Plots 

Bio-Engineering n  RQ: Test effectiveness of a Bio-feedback training method for proper manual wheel chair propulsion n  Ideal propulsion designed to decrease force and number of strokes ¡  Long smooth stroke during propulsive phase ¡  Allow the hand to drift down and back below the handrim during recovery phase 

Background n  Up to 70% of Manual Wheelchair Users report upper extremity pain often in the shoulder and wrist n  Teaching proper propulsion is not standardized n  Majority of the studies examine performance & technique at fixed speeds, on treadmills or dynamometers ¡  Have not used real life propulsion scenarios, overground ¡  Have not allowed participants to use their own wheelchairs 

Study Design n  Want to show improvement/growth ¡  Change over time, short and long term improvement n  Effectiveness of training across 2 other IVs ¡  Surface Type: carpet, tile, and ramp ¡  Speed: self-selected or targeted n  Design- Experiment ¡  Control Group ¡  Instruction Only Group n  Given a multimedia presentation showing proper form ¡  Bio-feedback Group n  Given the multimedia presentation n  Had a motor learning training program 

Design: True-Experiment 

Analysis n  Covariates to adjust for ¡  Weight – heavier you are more force used ¡  Time since injury – longer you have been injured, the more efficient pushing should be ¡  Level of injury – amount of function in wrists and hands n  Velocity is a confounder ¡  The faster you are going, the more force exerted and not a function of poor technique ¡  Velocity is not a constant like other covariates 

General Linear Mixed Model (GLMM) n  Multi-Level Model ¡  Can be used for repeated measure designs ¡  Measurement Occasions are nested within Individuals ¡  Like a more flexible mixed design ANOVA n  Can specify the covariance structure of the errors directly n  If the strict assumption of Compound Symmetry is violated- don’t have to use a correction n  Compound Symmetry: assumption that the measures at all time periods are equally related and the same for all groups n  Velocity entered as time-varying covariate ¡  Its value changes for each measurement occasion 

Study n  Sample ¡  n = 27 ¡  Had to use same wheelchair throughout study ¡  No difference between groups in demographics n  Four outcome measures ¡  Contact Angle (larger is better) ¡  Stroke frequency ¡  Peak Force (peak force during push) ¡  Rate of Rise (how rapidly force is applied to handrim, the quicker the better) 

Results, Simple Main Effects 

Results cont. 

Public Health n  RQ: Want to estimate number of Young Adults (18-29) in USA at risk for Type II Diabetes n  Why? Current screening methods for Diabetes (ADA) ¡  Over 45 and high BMI (≥26) ¡  Under 45, high BMI, and another other risk factor n  Diabetes Diagnosis ¡  Impaired Fasting glucose (IFG) ≥ 126mg/dl ¡  Prediabetes IFG ≥ 100mg/dl 

Metabolic Syndrome (MetS) n  MetS is one of the most reliable and best predictors of future Diabetes (65% of people with Diabetes had MetS) n  Criteria for MetS is 3 of the following è n  Having both IFG & 3/4 of other MetS risk factors increases chance of Type II Diabetes by twenty-fold n  Waist Circumference ¡  Men ≥ 102cm ¡  Women ≥ 88cm n  Triglyceride level ≥ 150mg/dl n  Cholesterol HDL ¡  Men ≥ 102mg/dl ¡  Women ≥ 88mg/dl n  Blood Pressure ¡  SBP ≥ 130mmhg ¡  DBP ≥ 85mmhg n  IFG ≥ 100mg/dl 

Study - NHANES n  Used data from National Health and Nutrition Examination Survey (NHANES) n  NHANES is longitudinal study that combines survey & physical data n  Specifically over-sampled minorities, as classically they are under represented in national data n  Individuals can be “weighted” so that collectively they represent the population of the USA from census 

Design & Analysis n  Stratified Cluster Sample, n ≈ 10,000 ¡  1,191 young adults in NHANES ¡  468 were included (1/2 missing lab data, pregnancy, etc) n  Design: Observational Study using Secondary data analysis ¡  Data set available online for free n  Analysis: Complex survey design was used, need to account for that when getting estimates n  Specifically take into account that clusters were stratified and individuals were weighted ¡  Clustering- saves money by choosing individuals close in location ¡  Stratifying- makes sure to get representation of everyone ¡  Weighting- to get sample to equal population according to census 

Results n  The bad news: ¡  Over half of adult youth are overweight n  BMI ≥ 26 ¡  24.9% are Obese n  BMI ≥ 30 ¡  IFG found in 28.2% ¡  16% had MetS, not with IFG ¡  6.4% had both IFG & MetS n  What about the lean? ¡  In our sample 12% of individuals had prediabetes (IFG between 100 and 125) but no other risk factor ¡  0% were n  Told they had pre- diabetes n  Warned by health care providers that they were at risk… 

Education – School Drop Outs n  RQ: Want to predict who is at risk for dropping out of School ¡  Collect validity evidence showing ADSI-E is effective in predicting drop outs n  High School drop outs are at risk for mental health problems, substance abuse, crime, etc n  If we could predict drop out and what factors influence it, educators could devote resources to help 

DUSI into ADSI-E n  The Drug Use Screening Inventory (DUSI) was adapted for school settings n  Converted to the Adolescent Development Screening Inventory for Education (ADSI-E) ¡  Minor changes in wording of items ¡  Items were grouped into different sub categories n  Validity evidence gathered over 15+ years for DUSI, since this study further evidence has been collected supporting use for the ADSI-E 

ADSI-E n  ADSI-E ¡  20mins to complete ¡  154 yes or no, 8 are lie scale ¡  1 Total and 9 domain scores n  Domains ¡  Peer Relationships n  Peer pressure ¡  School Adjustment n  School climate ¡  Family Systems n  Family background ¡  Leisure and Recreation n  Physical Activity ¡  Physical Health ¡  Emotional Health ¡  Social Competence n  Social skills ¡  Behavior Patterns n  Externalization of negative behavior ¡  Substance Abuse 

Study Design & Analysis n  Design: Quasi-experimental, naturally occurring groups ¡  Would be unethical to assign kids to drop out group n  Analysis: ANOVA, comparing groups on ADSI-E n  Sample drawn from Blue Water School district in Ontario n  Regularly attending students n = 442 n  “Drop-outs” n = 97 ¡  Were enrolled in one of three off-site regional Centers for Individual Studies ¡  Part-time students in alternative programs ¡  Conservative measure of “drop-out” as these were not students with 0 engagement 

Results n  The overall score for dropouts (M = 55.66, SD = 24.34) was significantly higher than the overall score for non-dropouts (M = 45.61, SD = 20.56), F(1, 537) = 17.73, p < .001 (p η2 = .03; d=.35) n  All domains but Emotional Health were significantly different n  Largest effect sizes: ¡  Substance Use p η2 = .08 d=.59 ¡  School Adjustment p η2 = .07 d=.55 ¡  Family Systems p η2 = .06 d=.51 

Results Table Domain Group N Mean Std. Deviation Physical Health Dropout 93 4.280 2.013 Non-dropout 436 3.454 1.937 Leisure & Recreation Dropout 94 5.649 1.916 Non-dropout 433 4.737 2.177 School Adjustment Dropout 91 9.813 4.297 Non-dropout 431 6.865 3.942 Family System Dropout 88 6.227 3.183 Non-dropout 430 3.984 3.292 Behaviour Patterns Dropout 85 9.082 3.840 Non-dropout 426 8.042 4.078 Social Competence Dropout 82 4.573 3.147 Non-dropout 416 3.726 2.718 Peer Relationships Dropout 85 7.835 2.927 Non-dropout 420 5.900 2.973 Emotional Health Dropout 82 8.585 3.617 Non-dropout 413 7.697 3.847 Substance Use Dropout 85 6.024 4.257 Non-dropout 406 3.192 3.491 

Social Work & Justice System n  RQ: Examine the patterns of Child-Welfare system involvement among youth, who have aged out of the child welfare system ¡  Look at the impact of Race n  Data from Department of Human Services (DHS) anyone involved with Child/Youth-Welfare n  All kids born between 1985-1994 in Allegheny County with system involvement n  42,735 youth from 23,754 families ¡  51% male ¡  47% Black, 45% White, 6% other n  Aged out youth: are kids who do not have a stated goal of reunification with their families prior to leaving the custody of the child welfare system 

Background n  Who qualifies? Any youth who ¡  Experienced a placement n  In Foster or Kinship Care n  In Group Home or Residential Facility ¡  Received any type welfare service n  Mental Health (MH) n  Drug & Alcohol (DA) n  Employment Training (ET) n  Housing & Homeless (HH) ¡  Involved with Justice System n  Juvenile Justice (JJ) n  Jail n  Special Subsample of Aging Out Youth: ¡  17 and older when left care with one year or more in out of home placement 

Study 1: Design & Analysis n  Design: Quasi-Experimental, uses Administrative Data but primary data collection n  Part 1: Two-step Cluster Analysis ¡  Identifies homogeneous subgroups ¡  Used the diversity of DHS services to create groups n  Part 2: Test Cluster differences on 12 variables using ANOVAs- Are clusters meaningful? n  Part 3: Test Mediation by Race- Can race explain the Cluster differences? n  Part 4: Test Cluster Differences using ANCOVA to adjust for potential confounders 

Study 1: Part 1 - Cluster Results 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Low involvement (17%) MH Only (30%) MH & DA (22%) MH & JJ (15%) MH & Jail (17%) Mental health Drug & alcohol Juvenile justice Jail 

Study 1: Part 2 – ANOVA Results n  The two low involvement groups compared to the high involvement groups ¡  Have more placements (5, 7) : (9, 11, 9) ¡  Spend larger % of time not in foster care (77%, 63) : (39%, 42, 45) ¡  More likely to have only foster care (52%, 29) : (7%, 11, 13) ¡  Less likely to run away (23%, 29) : (46%, 61, 50) n  All significant main effects, and post-hoc comparisons controlling for type I error inflation 

Study 1: Part 3 – Race Mediation n  Could race mediate (explain) these cluster differences? n  Why look at race? 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% Allegheny County Youth CYF Involved Aged Out African American White Cluster %AA Low Inv. 82% MH only 66% MH & DA 60% MH & JJ 71% MH & Jail 79% 

Study 1: Part 4 – ANCOVA Results n  Covariates ¡  Race, Age, Sex, ever Running Away and Age at first placement n  Ran an Exploratory Factor Analysis on MH services ¡  As a reduction technique, whether to look at all MH services collectively or individually ¡  Results indicated 1 underlying factor ¡  MH Factor score = sum of services received n  DVs ¡  Number of placements ¡  Years in Placements ¡  Mental Health Services ¡  Percent time out of home in Foster & Congregate Care 

Study 1: Take Home Message n  Numbers with subscripts differ significantly n  Low Involvement and MH only groups had less placements, more time in foster care, less time in Congregate Care ¡  Placement type more important than length n  Time spent in placement not as important as placement stability (p η2) Low Inv. MH only MH & DA MH & JJ MH & Jail # Placements .058 6.17 6.83 a 8.33 b 10.11 a 8.44 Years in Placement .013 4.42 a 5.10 4.36 4.47 4.32 Foster Care % .068 73 a 55 b 40 b 42 b 42 Cong. Care % .065 19 a 36 b 51 c 44 c 47 MH Services - 3.12 a 3.52 a 3.59 b 3.24 

Study 2: Juvenile Justice n  RQ: Examine how placement experiences and services received effected kids’ entrance into JJ n  Also to examine impact of Race and Sex on the relationship between experiences and JJ (Race & Sex as Moderators) n  Child welfare-involved youth are more likely than general population to become involved with in JJ system ¡  Timing is important for looking at what may affect kids’ entering the Justice system n  Only considered child welfare involvement if it happened before JJ 

Study 2: Analysis - GEE n  Our analysis has to account for the fact that siblings are more similar than non-siblings ¡  Otherwise it is a Violation of Independence n  Our outcome variable is dichotomous n  Generalized Estimating Equation used ¡  Provides unbiased marginal (population-average) regression coefficients regardless of the correlation structure of the errors ¡  Part 1: Entire sample (any interaction with DHS) ¡  Part 2: Just those who had placements 

Study 2: Demographics n  Used dichotomous variables, to indicate ever having the experience ¡  Conservative estimate of impact of the variables n  3,712 youth from 2,918 families n  AA and White only ¡  Other races too few n  20% JJ involvement n  55% had CC n  18% Ran away n  70% had MH n  25% had DA n  83% cases open after 13 n  Averaged 4.8 (SD = 4.5) placements n  Out of home 2.3 years (SD= 2.93) 

Study 2 : Part 2 –Results n  Placement Sample results ¡  Being African American increase Justice involvement ¡  Being Male increases it ¡  MH increases it ¡  DA protective ¡  Closing Case early protective ¡  Running away is protective? 

Study 1: Part 2 –Results: Odds Ratios Main Effect Model 95%CI Interaction Model 95%CI African American 1.90** 1.56-2.32 2.38** 1.43-3.98 Boys 3.35** 2.79-4.02 3.89** 2.39-6.34 Case open after 13 2.74** 2.00-3.77 2.95** 2.12-4.10 # of placements 1.11** 1.08-1.14 1.11** 1.08-1.14 Years in placement .92** .89-.96 .93** .89-.96 Congregate care 1.25* 1.01-1.55 3.60** 2.11-6.14 Runaway .74* .57-.95 .72* .56-.92 Mental health services 1.46** 1.17-1.81 1.47** 1.18-1.84 Substance abuse services .73* .59-.91 .73* .59-.90 AA & Male 1.64* 1.09-2.46 AA & congregate care .48** .32-.73 Male & congregate care .47* .30-.76 n  Interaction Terms- what do they tell us? 

Study 2: Part 3 – Predicted Probabilities n  African American boys are twice as likely as white boys to have JJ n  African American girls have only a 50% increased likelihood of JJ compared with white girls 0% 5% 10% 15% 20% 25% 30% 35% 40% Boys Girls African American White 

Study 2: Part 3 – Predicted Probabilities n  In general, AA more likely to have had CC than White Youth n  White Youth in CC more than 2x as likely to have JJ involvement 0% 5% 10% 15% 20% 25% 30% Yes No Congregate Care African American White 

Study 2: Part 3 – Predicted Probabilities n  On Average, boys more likely to have JJ involvement than females n  Girls who were in CC 2x as likely to have JJ involvement 0% 5% 10% 15% 20% 25% 30% Yes No Congregate Care Boys Girls