Get Them Out! Self-Administered IV Antibiotics at Home The - PowerPoint PPT Presentation

Get Them Out! Self-Administered IV Antibiotics at Home The Importance of a Propensity Score Robert W. Haley, MD Division of Epidemiology University of Texas Southwestern Medical Center Dallas No Disclosures

The Problem • Every hospital has patients who require long- term IV antimicrobial infusion. – Staph septicemia – Bacterial endocarditis – Diabetic foot – Osteomyelitis • After the initial workup, the patients occupy a hospital bed only to receive the infusion. • Insured or charity patients can be discharged to receive infusions in an LTAC or home health nurse, but uninsured patients stay in hospital.

Disadvantages of Long-term Antimicrobial Infusion in Hospital Burden on safety-net hospitals Burden on patients Risk of Intensity complications of services (e.g., infection) 3 Hospital days 42

Definition of OPAT • O utpatient P arenteral A ntimicrobial T herapy • Provision of IV antibiotic therapy in at least 2 doses on different days outside the hospital. • Goals – Allow patients to complete treatment safely and effectively in the comfort of their home or another outpatient site – Avoid the inconveniences, complications, and expense of hospitalization

Models of outpatient parenteral antimicrobial therapy (OPAT) delivery Paladino J A , and Poretz D. CID 2010;51:S198-S208

Definition of OPAT • O utpatient P arenteral A ntimicrobial T herapy • Provision of IV antibiotic therapy in at least 2 doses on different days outside the hospital. • Goals – Allow patients to complete treatment safely and effectively in the comfort of their home or another outpatient site – Avoid the inconveniences, complications, and expense of hospitalization • But not available to patients without funding (e.g., private insurance, Medicare, Medicaid or other local funding option).

Kavita Bhavan, MD

“ Let unfunded patients do it themselves” S-OPAT Self-Administered Outpatient Parenteral Antimicrobial Therapy

S-OPAT Program

Intervention • Developed program in 2009 as an alternative for uninsured patients to complete long-term antibiotic therapy at home comparable to services received in traditional funded settings. • Patients undergo bedside teaching and competency assessment prior to discharge from hospital. • Transitioned from the hospital into a dedicated post- discharge OPAT clinic, and followed weekly by nurses for PICC line care and at fixed intervals by physicians to assess clinical response to therapy.

Best Practice Methods • Dedicated multidisciplinary OPAT team: Physician, Pharm D, Care Management, RN • Effective multilingual patient education material at the appropriate level of health literacy and employ the “teach back method” for bedside teaching • Standardized core competency tool to test and record patient’s ability to self -administer IV antibiotics 11

Incorporating Patient Safety into Transition of Care

Coaching patients for successful outcomes

Patient Education

Best Practice Methods Teach-Back: Closing the Loop Schillinger D, Piette J, Grumbach K, Wang F, Wilson C, Daher C, Leong-Grotz K, Castro C, Bindman A. Closing the Loop Physician Communication With Diabetic Patients Who Have Low Health Literacy. Arch Intern Med/Vol 163, Jan 13, 2003

Testing for Competency

Teaching Tools

Specific Instructions

Preparing Antibiotics

Infusion by Gravity Wire coat hanger

Weekly Followup in S-OPAT Clinic

Study to Evaluate S-OPAT Objective Determine whether indigent, often poorly educated and mostly non-English-speaking patients can self-administer long-term IV antibiotics at home (S-OPAT) as safely and effectively as traditionally accepted models of outpatient care by a healthcare practitioner available to patients with funding (H-OPAT)

Outcomes • Compared patients treated in S-OPAT with those treated in H-OPAT on 2 outcomes – 30-day readmission rate – 1-year mortality rate • Calculated total number of hospital bed days avoided, as reflected by number of days a patient self-administered parenteral antibiotic therapy as an outpatient under the S-OPAT program

Controlling for Selection Bias • An observational study (non-randomized) • Patients in the S-OPAT and H-OPAT groups differed on several important measures. – Healthcare funding (insurance, Medicare, Medicaid) – Language – Nationality and US citizenship – Educational level – Age • These differences created a strong potential for selection bias in the outcome. • Must control for this in the analysis.

Two Approaches in a Multivariable Logistic Regression Analysis • Enter Covariates into the multivariable analysis • Enter a Propensity Score into the multivariable analysis

Two Approaches in a Multivariable Logistic Regression Analysis • Enter Covariates into the multivariable analysis – Age – Sex – Race – Country of origin – Source of payment – Education level – Income • Enter a Propensity Score into the multivariable analysis

Two Approaches in a Multivariable Logistic Regression Analysis • Enter Covariates into the multivariable analysis – Age Covariates control – Sex confounding, but not – Race necessarily selection bias. – Country of origin – Source of payment – Education level – Income • Enter a Propensity Score into the multivariable analysis

Two Approaches in a Multivariable Logistic Regression Analysis • Enter Covariates into the multivariable analysis – Age Covariates control – Sex confounding, but not – Race necessarily selection bias. – Country of origin – Source of payment – Education level – Income • Enter a Propensity Score into the multivariable analysis A propensity score measures each patient’s propensity (probability) of receiving the treatment (S-OPAT) based on their characteristics.

Propensity Score • Definition: An individual patient’s probability of receiving the treatment conditional on measured covariates. – “How do you develop a propensity score?” R. Haley, Epi for Clin Investigator

Development of Propensity Score – Instead of putting covariates into the main multivariable logistic regression model: Outcome = Treatment Covar-1 Covar-2 . . . Covar-I (output is the odds ratio) – Develop a logistic regression model of the treatment: Treatment = Determinant-1 Determinant-2 . . . Determinant-I / pred=PS And output the probability of treatment conditional on the determinant variables. This is the PS (a continuous variable = probability of treatment) – Within the strata of the PS, the probability of getting the treatment is the same. R. Haley, Epi for Clin Investigator

Development of Propensity Score – Instead of putting covariates into the main multivariable logistic regression model: Outcome = Treatment Covar-1 Covar-2 . . . Covar-I PS (output is the odds ratio) – Develop a logistic regression model of the treatment: Treatment = Determinant-1 Determinant-2 . . . Determinant-I / pred= PS And output the probability of treatment conditional on the determinant variables. This is the PS (a continuous variable = probability of treatment) – Within the strata of the PS, the probability of getting the treatment is the same.

4 Ways to Use the Propensity Score • PS Covariate Adjustment (most used): – Introduce the PS into the multivariable logistic model of outcome as a new covariate. • PS Stratification: – Stratify the outcome analysis on the PS. • PS Matching: – Match • PS Weighting: – Weight the multivariable logistic model of outcome with the inverse PS. R. Haley, Epi for Clin Investigator

Summary of Patient Selection a b a Patients who were homeless, had a history of IV drug abuse, or were medically unstable b The eligibility criteria are given in the appendix of the paper.

Outcome variable is S-OPAT vs H-OPAT Logistic Regression Model of S-OPAT vs H-OPAT to Develop the Propensity Score

Multivariable Proportional Hazards Regression Model of 30-day Readmission Model 1 controls for confounding by entering covariates into the model.

Multivariable Proportional Hazards Regression Model of 30-day Readmission The OR of 0.59 indicates that S-OPAT had a 41% lower 30- 1 – 0.59 = 41% reduction day readmission rate than H-OPAT. % reduction = 1 - OR Model 1 controls for confounding by entering covariates into the model.

Multivariable Proportional Hazards Regression Model of 30-day Readmission 1 – 0.59 = 41% reduction Model 1 controls for confounding by entering covariates into the model. Model 2 also controls for selection bias by entering the propensity score into the model.

Multivariable Proportional Hazards Regression Model of 30-day Readmission 1 – 0.59 = 41% reduction 1 – 0.53 = 47% reduction Model 1 controls for confounding by entering covariates into the model. Model 2 also controls for selection bias by entering the propensity score into the model.

Multivariable Proportional Hazards Regression Model of 1-Year Mortality 1 – 0.94 = 6% reduction 1 – 0.86 = 14% reduction Model 1 controls for confounding by entering covariates into the model. Model 2 also controls for selection bias by entering the propensity score into the model.