Comparative Effectiveness of Rehabilitation Services for Survivors - - PowerPoint PPT Presentation

ANNUAL MEETING

• OCT. 31-NOV. 2, 2017

ARLINGTON, VA

#PCORI2017

Comparative Effectiveness of Rehabilitation Services for Survivors

• f an Acute Ischemic Stroke

Janet Prvu Bettger, ScD, FAHA

Associate Professor, Duke University @jpbettger November 1, 2017 On b n beha ehalf o

• f: Laine Thomas, PhD; Ying Xian, MD, PhD; Li Liang, MS; Jerome J.

Federspiel, MD, PhD; Laura E. Webb, BS; Cheryl D. Bushnell, MD, MHS; Pamela W. Duncan, PT, PhD; Eric D. Peterson, MD, MPH; Lee H. Schwamm, MD; Joel Stein, MD; Gregg C. Fonarow, MD; Helen Hoenig, MD, MPH; Cris Montalvo, MA; Mary

• G. George, MD, MSPH; Barbara J. Lutz, PhD, RN

ANNUAL MEETING | #PCORI2017

2

Why Methods Matter

• Putting Research to Work for Individual Patients
• Using instrumental variable (IV) methods to improve

understanding of treatment effects for survivors of acute ischemic stroke

ANNUAL MEETING | #PCORI2017

Specific Aim and Approach

Aim: To compare care provided in inpatient rehabilitation facilities (IRF) and skilled nursing facilities (SNF) on ischemic stroke patient outcomes Data Sources: American Heart Association (AHA) Get With The Guidelines- Stroke registry linked with Medicare Claims Outcomes: mortality post-discharge, a composite of all-cause rehospitalization or death, days at home Analyses:

• Proportional hazards regression with inverse propensity weighting to

adjust for measured differences in demographic, clinical and hospital characteristics.

• Instrumental variable approach to adjust for both measured and

potential unmeasured confounding

• IV1 = hospital-specific proportion of patients discharged to IRFs vs SNFs
• IV2 = differential distance

ANNUAL MEETING | #PCORI2017

Importance of comparative effective research

• Post-acute care (PAC) is the largest driver of variation in

Medicare spending

• Stroke patients are the highest users of PAC
• More than half of stroke patients are discharged from the

hospital to receive PAC in an IRF or SNF

• Discharge planning and decision making is highly variable

ANNUAL MEETING | #PCORI2017

IRFs and SNFs are Different Interventions*

Requirements IRFs SNFs Therapy 3 hours/day No requirement MD Supervision Daily, direct Not daily RN - Nursing 24 hours/day 8 hours/day Length of Stay 16 days (mean) 35 days (mean) Licensed As a hospital As a nursing home Standards Intense, multi- disciplinary, coordinated care Few regulations guiding provision of care

*PCORI study used an intention-to-treat (ITT) analysis and did not assess potential differences in delivery of the interventions.

ANNUAL MEETING | #PCORI2017

Exclusion Criteria

Individuals who were not suitable candidates for IRF or SNF care and those who would or would not experience the outcome regardless of intervention Of 76,315 ischemic stroke patients, we excluded patients with:

• CMS index discharge destination of hospice (community or inpatient)

(N=159)

• GWTG comfort care measures (earliest documentation of comfort measures
• nly at day 1 or 2) (N=620)
• Metastatic cancer documented as a co-occurring condition during index

stroke admission

• Evidence of LTC in the 6 months prior to index admission (N=562)
• Acute care LOS>14 days (N=5,000)
• From hospitals with >25% missing on past medical history variables (N=358

patients, 24 hospitals)

ANNUAL MEETING | #PCORI2017

Patient Sample for IRF/SNF Study

69,004 acute ischemic stroke patients who received acute care 1,147 U.S. hospitals Inpatient Rehabilitation Skilled Nursing 34,574 (50.1%) 34,430 (49.9%)

ANNUAL MEETING | #PCORI2017

Baseline Characteristics of Both Groups

Demographic Variables IRF SNF

• Std. Difference*

Age mean (std) 78.7 (7.5) 83.0 (7.5) 58.0% Female sex No (%) 19,621 (56.8) 23,186 (67.3) 22.0% Race No (%) Hispanic 1,331 (3.9) 986 (2.9) 7.0% Black 3,416 (9.9) 3,132 (9.1) American Indian or Alaska Native 70 (0.2) 34 (0.1) Asian 483 (1.4) 457 (1.3) Caucasian 28,136 (81.5) 28,619 (83.3) Native Hawaiian or Pacific Islander 63 (0.2) 46 (0.1) UTD 1,043 (3.0) 1,070 (3.1) Caucasian race No (%) 28,168 (81.5) 28,705 (83.4) 5.0% Health Service Use 6 months Prior # hospitalizations in 6mon prior to index 0.3 (0.7) 0.5 (0.8) 22.9% # hospitalizations in 6mon prior to index=0 27,087 (78.5) 23,353 (68.3) 24.2% # hospitalizations in 6mon prior to index=1 5,476 (15.9) 7,356 (21.5) # hospitalizations in 6mon prior to index>=2 1,924 (5.6) 3,470 (10.2) IRF adm in 6mon prior to index 935 (2.7) 655 (1.9) 5.3% SNF adm in 6mon prior to index 1,300 (3.8) 5,718 (16.6) 43.5%

* Standardized difference = difference in means or proportions divided by standard error; imbalance defined as absolute value greater than 0.20 or 20% (small effect size)

ANNUAL MEETING | #PCORI2017

Baseline Characteristics of Both Groups

* Standardized difference = difference in means or proportions divided by standard error; imbalance defined as absolute value greater than 0.20 or 20% (small effect size) Medical History IRF SNF

• Std. Difference*

AMI/CAD 11,332 (32.8) 11,953 (34.7) 4.1% Chronic HF 3,274 (9.5) 4,513 (13.1) 11.5% Diabetes 11,395 (33.0) 10,957 (31.8) 2.4% Peripheral vascular disease 4,221 (12.2) 4,242 (12.3) 0.3% Stroke 11,181 (32.3) 12,931 (37.6) 11.0% Atrial fibrillation 7,671 (22.2) 9,844 (28.6) 14.8% Carotid Stenosis 1,676 (4.8) 1,599 (4.6) 1.0% Dyslipidemia 13,865 (40.1) 11,586 (33.7) 13.4% Hypertension 27,178 (78.6) 27,092 (78.7) 0.2% Smoking 3,931 (11.4) 2,442 (7.1) 14.8% Prosthetic Heart Valve 510 (1.5) 480 (1.4) 0.7% Cancer 1,485 (4.3) 1,746 (5.1) 3.7% COPD 5,964 (17.2) 6,807 (19.8) 6.5% Dementia 1,404 (4.1) 4,813 (14.0) 35.2% Liver disease 199 (0.6) 186 (0.5) 0.5% Peptic ulcer disease 414 (1.2) 509 (1.5) 2.4% Renal disease 3,963 (11.5) 5,238 (15.2) 11.1% Rheumatic disease 984 (2.8) 947 (2.8) 0.6% Disability prior to admission 16,302 (47.2) 14,176 (41.2) 12.1% Medications prior to admission No (%) Antithrombotic (Anticoagulation or Anti 19,420 (56.2) 19,496 (56.6) 0.9% Antihypertensive 25,690 (74.3) 26,114 (75.8) 3.6% Cholesterol reducer 13,718 (39.7) 11,938 (34.7) 10.4% Diabetic medications 8,073 (23.3) 7,724 (22.4) 2.2%

ANNUAL MEETING | #PCORI2017

Baseline Characteristics of Both Groups

Documented in Acute Care IRF SNF

• Std. Difference*

EMS from home/scene 26,004 (75.2) 28,222 (82.0) 16.5% NIHSS mean (std) 8.0 (6.6) 9.5 (7.9) 21.2% tPA in this hospital 2,539 (7.3) 1,361 (4.0) 14.7% Any serious complications of tPA 153 (0.4) 165 (0.5) 0.5% Stroke unit 19,150 (55.4) 16,759 (48.7) 13.5% DVT prophylaxis by hospital day 2 25,819 (74.7) 25,114 (72.9) 3.9% Ambulate independently at discharge 6,594 (19.1) 5,675 (16.5) 6.8% LOS mean (std) 5.3 (2.7) 6.1 (2.9) 29.6% Medications/tx at discharge No (%) Antithrombotic (Anticoagulation or Antiplatelet) 32,601 (94.3) 30,975 (90.0) 16.1% Antihypertensive 28,018 (81.0) 27,571 (80.1) 2.4% Cholesterol reducer 23,467 (67.9) 18,594 (54.0) 28.7% Smoking cessation 4,564 (13.2) 3,084 (9.0) 13.6% Hospital characteristics No (%) Number of Beds 466 (292.6) 420 (287.7) 15.7% Teaching hospital 21,647 (62.6) 19,904 (57.8) 9.8% Rural 1,194 (3.5) 2,246 (6.5) 14.1% Certified PSC 18,833 (54.5) 17,441 (50.7) 7.6% Hospital level composite of QOC measures 0.9 (0.1) 0.9 (0.1) 11.0%

Additional factors examined: SES at neighborhood level; in-hospital indicators of patient complexity (devices to support breathing or feeding, catheter use, ICU stay, palliative care), complications during hospital admission (cardiac/respiratory arrest, VT/PE, renal failure, sepsis, decubitus ulcer, UTI)

* Standardized difference = difference in means or proportions divided by standard error; imbalance defined as absolute value > than 0.20 or 20% (small effect size)

ANNUAL MEETING | #PCORI2017

Patients Discharged to an IRF vs. SNF are Different

Compared with IRF patients, SNF patients: Are older More are female Were hospitalized more in the prior 6 mo. Were in a SNF in the prior 6 mo. More have dementia Have more severe strokes Had a longer hospital LOS

ANNUAL MEETING | #PCORI2017

Factors considered for treatment selection that may be associated with the outcome

What we know

• Stroke severity
• Ability to ambulate
• Medicare FFS
• Facility availability

What we don’t know

• Physical & cognitive function
• Able to participate in 3 hours

daily rehab

• Social support

Unmeasured confounders Measured confounders

ANNUAL MEETING | #PCORI2017

Instrumental Variables (IV) to Address Confounding

IV 1: Practice Patterns

Proportion of patients in a hospital who are discharged to an IRF rather than a SNF (calculated as % IRF) Proxy for potential hospital-level unmeasured confounding

IV 2: Differential Distance

The distance between the patient’s residence and the nearest IRF minus the distance between the patient’s residence to the nearest SNF Proxy for potential patient-level unmeasured confounding

ANNUAL MEETING | #PCORI2017

Testing IV Assumptions: Hospital Practice Patterns for Discharging Patients to Either IRF or SNF

Percent of Patients Discharged to an IRF

ANNUAL MEETING | #PCORI2017

Testing IV Assumptions: Practice Patterns

• %IRF extremely strong relationship to actual treatment selection
• Effect estimate is likely to have good precision and to be representative of a

very large portion of the total population

• %IRF is moderately associated with measured patient and hospital characteristics

that would suggest a possibility for unmeasured confounding in favor of IRF.

ANNUAL MEETING | #PCORI2017

Testing IV Assumptions: Differential Difference

Probability of Receiving IRF Positive values indicate SNF closer to patient than IRF

ANNUAL MEETING | #PCORI2017

Testing IV Assumptions: Differential Difference

• Association of DD with actual treatment selection is not nearly as strong
• DD achieved very good balance on measured patient characteristics and reasonable balance
• n hospital characteristics
• Suggests that the DD IV can minimize unmeasured confounding (particularly factors

related to patient selection) and may provide unbiased results

• IV analysis based on DD would be expected to have less precision, and wider confidence

intervals.

• Results would represent a unique population whose choice of rehab setting is

influenced by distance.

ANNUAL MEETING | #PCORI2017

18

Results: Effectiveness of Care In IRF vs SNF

ANNUAL MEETING | #PCORI2017

19

Summary Comments on Results

• All of the methods indicate short term mortality

benefit for patients treated at IRF vs. SNF

• RR are similar for %IRF IV compared to distance IV
• Confidence intervals for %IRF IV are much smaller
• Precision will be important for studying HTE: Focus on

%IRF IV in subsequent analyses

ANNUAL MEETING | #PCORI2017

20

HTE

“the nonrandom, explainable variability in the direction and magnitude of treatment effects for individuals within a population” Our approach

• Identify the most likely clinical effect modifiers: age, stroke

severity, predicted mortality (as a summary of overall risk)

• Apply the same IV methodology using %IRF IV

ANNUAL MEETING | #PCORI2017

21

Subgroup Analysis for 90-day Outcomes

ANNUAL MEETING | #PCORI2017

22

Subgroup Analysis for 1-year Outcomes

ANNUAL MEETING | #PCORI2017

23

Study Limitations (regardless of HTE)

• Concern about unmeasured confounding
• Limitations of the available instrumental variables
• No falsification endpoints could be identified to help

validate success of adjustment

• No clinical trial data exist (for any aspects of this

analysis) to help validate success of adjustment

• Sample limited to Medicare fee-for-service, adults > 65

years, treated at GWTG-Stroke participating hospitals

ANNUAL MEETING | #PCORI2017

24

Our Team

Stroke Survivors and Caregivers Janet Bettger Eric Peterson Ying Xian Jeff Federspiel Judy Stafford Jinjing Wu Laura Webb Li Liang Laine Thomas Pam Duncan Cheryl Bushnell Mary George, CDC (State Health Depts/Coverdell) Lee Schwamm, Mass General (Neurology/GWTG) American Heart Association Staff & Volunteers Barbara Lutz, UNC-W (Rehab & Community Health Nursing) Joel Stein, Columbia (PM&R/NECC) Helen Hoenig, VA (PM&R/Gero/VA) Cris Montalvo, Duke (SLP / Survivor Support)

ANNUAL MEETING | #PCORI2017

25

Acknowledgments

• To PCORI for recognizing the importance of this work by funding us to move the

science forward that will change practice

• To the American Heart Association’s GWTG-Stroke Science Subcommittee for the

input

• This research team for their rigor and support
• The advisory group for looking at the issues and the data from so many different

perspectives

• The stroke survivors and caregivers for sharing their stories to guide us to ask

research questions that are meaningful to them

ANNUAL MEETING | #PCORI2017

26

Thank you!

Janet Prvu Bettger, ScD, FAHA

Associate Professor, Duke University @jpbettger janet.bettger@duke.edu November 1, 2017