Health Care Costs A Treatment-Effects Approach Marina Soley-Bori, - - PowerPoint PPT Presentation

www.rti.org

RTI International is a registered trademark and a trade name of Research Triangle Institute.

The Influence of Relational Climate on Health Care Costs

A Treatment-Effects Approach

Marina Soley-Bori, PhD

1

Introduction: Diabetes care

2

Social worker Primary care provider Registered nurse Licensed nurse Nutritionist Psychologist Pharmacist

• Complex to manage (5 comorbidities on average)
• Low guideline compliance
• Twice the costs of non-diabetic patients

Does team functioning influence costs of diabetes care?

Beasley et al. 2004, American Diabetes Association 2008, U.S. Department of Veterans Affairs,2014 , Bojadzievski &Gabbay 2011.

Introduction: Relational Climate

3

• Shared perceptions of interpersonal interactions

including teamwork, conflict resolution, and diversity acceptance

• Measurement: VA All-Employees Survey (AES)
• A spirit of cooperation and teamwork exists in my

group

• Disputes or conflicts are resolved fairly in my work

group

• Differences among individuals are respected and

valued in my work group

• Relational climate is associated with better quality of

diabetes care, but what about costs?

Schneider, Benjamin et al. 2010, Mossholder, Richardson and Settoon 2011, J. K. Benzer et al. 2011, Wagner 2000

Study Goal

4

To assess the influence of relational climate on health care costs incurred by diabetic patients, accounting for the endogeneity in quality of care

Methods: Data sources and study sample

5

• Setting: Veterans Health Administration
• Data sources: VA administrative datasets

and the All Employees Survey

• Retrospective longitudinal study (2008-

2012)

• Inclusion criteria: At least 2 diabetes

diagnoses (ICD-9 250 250.93) each year between 2008-2012

• Assignment of patients to the clinic visited

most often based on primary care visits

6

Methods: Costs incurred by diabetic patients

Costs incurred by diabetic patients:

• VA Managerial Cost Accounting national

data abstract

• Differentiated among outpatient,

inpatient and total costs

• Adjusted for inflation: Producer Price

Index for General Medical and Surgical Hospitals (Bureau of Labor Statistics)

• Adjusted of regional labor cost

differences: CMS Medicare wage index adjusted for VA market areas (Wagner, 2015)

Cost models adjusted for:

• Patient characteristics:

 Age, gender, marital status, access priority status  Elixhauser index, Nosos risk score, insulin, mental health diagnosis  Quality of diabetes care (all-or- none process indicator)

• Clinic characteristics:

 Relational climate, urban/rural

• Parent facility characteristics:

 Teaching status

7

Methods: Modeling strategy (I)

• 1. Longitudinal data (unbalanced panel): Random vs.

fixed effects?  Hausman test: Ho: Error not correlated with regressors (RE preferred model) p<.001  clinic and year FE

• 2. Clustering of individuals assigned to the same clinic

 Correction in standard errors (robust covariance estimator)

• 3. Skewed distribution of costs

 Data cleaning: top-coding, deletion of implausible values and absolute studentized residual above 3.5  Generalized Linear Model (GLM) with the Gamma distribution and the Log link

• 4. Endogeneity of quality of diabetes care

 Treatment-effects model

$6,877 $6,786 $7,212 $7,803 $8,637

$5,000 $6,000 $7,000 $8,000 $9,000 2008 2009 2010 2011 2012

Evolution of Median Total Costs

8

Methods: Modeling strategy (II)

Relational Climate Costs incurred by diabetic patients

Three treatment-effects models with inverse probability weighting regression adjustment estimator

• Main equation: Cost model (GLM Gamma distribution and log link)
• Treatment equation: All-or-none model (logistic regression)
• Assumptions:

 Conditional independence assumption

• Standardized differences, density plots, assessment of the evolution of

the study variables across time by treatment group  Sufficient overlap assumption

• Probability distribution of guideline compliance among compliant and

non-compliant individuals  Double-robust property

• C-statistics, Hosmer-Lemeshow Test

Results: Descriptive Statistics (I)

9

Study sample: 1,568,180 patient observations across the study period. In a typical year, 200 clinics, 100 parent facilities and 300,000 patients Age (years) 65.5 (11.2) Male 96.4% (96.4-96.5) Marital Status

• Married

35.5% (35.4-35.7) Elixhauser Index

• 2-4 (both included)

80.7% (80.5-80.8) Mental Health Diagnosis

• Yes

37.9% (37.8-38.1) Insulin

• Yes

59.1% (58.9-59.3)

2010 data. N= 327,805 Notes: Mean for continuous variables (SD); percentage for categorical variables (95% CI)

Results: Descriptive Statistics (II)

10

$6,877 $6,786 $7,212 $7,803 $8,637 $5,000 $6,000 $7,000 $8,000 $9,000 2008 2009 2010 2011 2012

Evolution of Total Costs (median)

10.9 11.2 11.5 11.7 10.8 10.0 10.5 11.0 11.5 12.0 2008 2009 2010 2011 2012

Evolution of Relational climate (average)

51.6 56.5 56.8 55.9 55.5 48 50 52 54 56 58 2008 2009 2010 2011 2012

Evolution of the All-or-None Indicator (% Yes)

Results: The influence of relational climate on costs

11

Cost item MR (95% CI) 2012 Cost change per patient of 1 additional RC unit Outpatient Q=1 0.995*** (0.994,0.997)

• $64.93

Q=0 non-significant cost-neutral Inpatient Q=1 0.986*** (0.982,0.991)

• $569.16

Q=0 0.983*** (0.979,0.986)

• $468.72

Total costs Q=1 0.989*** (0.987,0.991)

• $109.95

Q=0 0.993*** (0.991,0.995)

• $67.97

Notes: MR=Mean Ratio, Q=Quality of diabetes care (measured by the all-or-none process indicator), Q=1 means guideline compliant, Q=0 means non guideline compliant. Cost models also adjusted for age, gender, marital status, enrollment priority, nosos risk scores, mental health diagnosis, insulin, rural/urban, the wage index, and teaching status.

Conclusions: Limitations

12

1. Costs incurred by diabetic patients vs. diabetes-related costs 2. Did not account for Medicare services used by dually eligible veterans 3. Partial modeling of the selection of patients into testing 4. Did not account for the cost of increasing relational climate by one unit 5. VHA population specific results

Conclusions/Discussion

13

• 1. Relational climate contributes to lower outpatient costs (only

among compliant individuals), lower inpatient costs, and lower total

• costs. The magnitude of the effect is modest though.
• 2. Future work should measure relational climate in other settings
• ther than primary care (e.g. emergency room) and assess how it

influences patient outcomes

• 3. More than 50% of the clinics had a relational climate score lower than

11.2 (max=15)  How do we change relational climate?

• Implementation of the Civility, Respect, and Engagement in the

Workforce (CREW) intervention at the VHA

More information

Marina Soley-Bori, PhD Research Economist RTI International msoleybori@rti.org

14

Questions?

Assumptions:

• Conditional independence

assumption (balance in the model covariates)

• Sufficient overlap assumption
• Double-robust property

(specification of the treatment equation)

• C statistic, Hosmer-Lemeshow

test (with bootstrapping).

• Prebigon link test (with

bootstrapping) Standardized differences (lower than 0.1) and variance ratios (closer to 1)

• C statistic= 0.69
• 86% p-values of H-L >0.05
• 88% p-values of P-L >0.05

Assumptions of the treatment-effects model