Three Essays on Physician Prescribing Behavior



Download 27,49 Kb.
Date conversion07.01.2017
Size27,49 Kb.

Three Essays on Physician Prescribing Behavior

  • Brian K. Chen
  • Orals Examination
  • December 4, 2006
  • Haas School of Business
  • University of California at Berkeley

Motivation Prescription drug expenditures: the fastest growing component of health care expenditures

    • 1990s: Prescription drug expenditures grew by 5% to 23% annually in most industrialized countries
      • United States: Fastest growing component of $1.9 trillion health care industry
      • In 2004: Third largest component in the US health care expenditures at 9%, following hospital (31%) and physician (22%) services
      • In 2002: Drug expenditures up by 15.3%, outstripping expenditures in hospital (9.5%) and physician (7.7%) services
    • By 2001: US$607 billion spent on prescription drugs worldwide
      • In nominal terms, top 20 GDP in the world
    • New drugs explain up to 40% of annual drug expenditures growth

Dissertation Outline and Research Questions

  • Chapter 1: What are the determinants in the adoption decision of new drugs?
    • Do physician, patient, and hospital characteristics matter in the likelihood/rate of adoption of new drugs?
  • Chapter 2: What is the health outcome impact of new drugs?
    • Do new drugs lead to better health outcomes?
  • Chapter 3: If high costs to patient affect drug use, do physicians take patient costs into consideration?

Why are these questions important?

  • Numerous implications
    • Who gets new drugs? Who prescribes new drugs?
      • Theoretical interest – consistent early adopters?
      • Policy interest
      • As first stage analysis for second essay
    • Are new drugs worth their cost?
      • If yes, what are the cost savings? How to encourage appropriate use of new drugs
      • If no, what are the additional costs compared to older, just-as-effective drugs?
    • If financial burden prevents access to new drugs, do physicians take this into consideration?
      • If only marginal improvement, physicians should prescribe older drugs to the financially burdened
      • If substantial improvement, implications for drug copayment policies

Contribution

  • Chapter 1:
    • Very little known about physician adoption of new drugs
    • But I need: theoretical framework?
  • Chapter 2:
    • No strong empirical evidence on the effectiveness of new drugs versus older drugs that corrects for selection bias
    • But I need: INSTRUMENT for treatment

Background

  • Quick statistics
    • Land Area: 13,823 square miles
    • Population (2006): 23,000,000
    • 2005 GDP: $U.S. 611.5 billion ($U.S. 326.5 billion)
    • 2005 Per Capita GDP: $U.S. 26,700 ($14,200)
  • Health Care in Taiwan:
    • 2003: $U.S. 11 billion
    • National Health Insurance, virtually 100% coverage
    • 5.7 hospital beds per 1,000 people, 1.4 physicians trained in Western medicine for every 1,000

Salient Features of Taiwan’s Health Care System

  • Closed System
  • Freedom of Choice
  • Lack of system of referrals
  • Commingling of diagnostic and dispensing services

◄Chapter 1► Adoption/Diffusion of New Therapeutic Agents

Research Question

  • Do patient, physician, and hospital characteristics matter in the likelihood/speed of adoption of new therapeutic agents?
    • Dependent variable: dichotomous variable indicating adoption of a new drug; or time to first adoption
    • Independent variables:
      • Patient Characteristics: age, gender, ER visits, hospitalizations (condition duration)
      • Physician Characteristics: type of doctor, age, gender, experience, tenure, past prescription pattern
      • Hospital Characteristics: Academic, urban, hospital/clinic
      • Drug Characteristics: New mechanism? Long or short-term treatment?
  • Are predictions based on these characteristics consistent across drugs?

Literature Review: Adoption of New Drugs

  • “Epidemic” studies
    • Menzel (1955), Coleman (1957), Peay (1988), Denig (1991) Nair (2006) (related)
  • “Firm Heterogeneity” studies
    • Steffesen (1999), Tamblyn (2003), Dybahl (2005)
  • Bayesian model of adoption
    • Coscelli (2004)

Motivation to Prescribe

  • Firm heterogeneity model: there exist characteristics that predict adoption of new technology
    • What these characteristics are remains an open empirical question
    • Are these characteristics constant across new drugs?
  • Patient Demand
  • Marketing Activities

Conceptual Framework: Predictions

  • Physician characteristics:
    • Prime age  greater adoption
    • Gender  unclear, general view is male  greater adoption
    • Past practice volume  greater adoption?
    • Type of doctor  family practice  less adoption
    • *Past use of drugs manufactured by same company  greater adoption
  • Patient characteristics
    • Age  unclear; depends on drug
    • Gender  unclear
    • *Higher Education  greater adoption
    • Condition severity  unclear
  • Hospital characteristics
    • Academic  greater adoption
    • Urban  greater adoption
    • Family practice  less adoption at hospitals
  • *Drug characteristics

Top Prescription Drugs in Taiwan by Sales, 2004

Top ICD-9-CM codes in Taiwan

Drugs introduced between 1997-2004

  • Atorvastatin (Lipitor) (19114/2097)
    • Date of introduction: November 1, 2000
    • Therapeutic class: statins
    • Indication: to lower cholesterol and thereby reduce cardiovascular disease.
    • With 2005 sales of US$12.2 billion under the brand name Lipitor, it is the largest selling drug in the world
  • Rosiglitazone (Avandia) (15281/1052)
    • Date of introduction: March 1, 2001
    • Therapeutic Class: thiazolidinedione
    • Indication: Anti-diabetic drug (Diabetes Type II)
  • Clopidogrel (Plavix) (7378/728)
    • Date of introduction: January 1, 2001
    • Therapeutic Class: Antiplatelet agent
    • Indication: is a potent oral antiplatelet agent often used in the treatment of coronary artery disease, peripheral vascular disease, and cerebrovascular disease.
    • In 2005 it was the world's second highest selling pharmaceutical with sales of US$5.9 billion

Other new drugs

    • Celecoxib (Celebrex)
      • Arthritis/Pain (April 1, 2001) (but: side effects) (15574/3952)
    • Esomeprazole (Nexium)
      • Heartburn/Acid Reflux (January 1, 2002) (4250)
    • Olanzapine (Zyprexa)
      • Schizophrenia/Bipolar (February 1, 1999) (5284)
    • Venlafaxine (Effexor)
      • Antidepressant ( October 1, 2000) (2296)
    • Montelukast (Singulair)
      • Asthma (July 1, 2001) (2489)
    • Quetiapine (Seroquel)
      • Schizophrenia/Bipolar (April 1, 2000) (2795)

Disease Code Combinations only < 1% of visits have no ICD9 code

Description of Data

  • Panel Data
    • Eight years of complete medical claims data for a random selection of 200,000 individuals from Taiwan’s population of 23 million
    • HOSB, PER, DOC and ID files
    • The age, gender, and expenditures of the randomly selected individuals do not differ significantly from the population
  • Time Series (Random Subsamples)
    • Outpatient Expenditures
    • Inpatient Expenditures
    • Prescription Drugs at Contracted Pharmacies (complete)

Summary Statistics - Hypertension

Empirical Strategy – Likelihood of adoption

  • Probit/Logit Model
    • Pat: Patient Characteristics: age, gender, past number of visits, ER visits, hospitalizations, multiple conditions?
    • Phys: Physician Characteristics: age, gender, experience, tenure, past prescription pattern
    • Hosp: Hospital Characteristics: Academic, urban, family practice
    • Endogenous variable? Omitted variables (Neglected heterogeneity)? New diagnoses?

Empirical Strategy – Duration to Adoption

  • Right-Censored Duration Model
    • Continuous or Discrete Time-Scaling
    • Nonparametric or parametric functional form?:
      • Weibull (increasing)
      • Log-logistic
    • Effect of Covariates (same as previous slide)
      • Proportional Hazard (+coeff  - hazard / +duration)
      • Accelerated Lifetime Hazard (1 unit +coeff  % +duration)
    • Other issues
      • Multiple spells? Time-varying covariates? (move from one hospital to another?) Unobserved Heterogeneity?

Diffusion pattern Lipitor (for Hypertensive Patients Only)

Preliminary Results – Panel Data Likelihood of Lipitor Adoption

Preliminary Results – “Pooled” Data Likelihood of Lipitor Adoption

Preliminary Results – Year by Year Likelihood of Lipitor Adoption

Discussion

  • Need to reconstruct data from scratch
  • Different types of severity
    • multiplicity of conditions, or severity of a single condition
  • Not surprising:
    • academic, urban providers more likely to adopt, patients with multiple indications more likely to be given Lipitor
  • A little surprising?
    • Female physicians more likely to adopt (probably problem from merged data); female patients more likely to receive
  • Quite surprising?
    • More serious patients less likely to be given Lipitor

Future Agenda

  • Better understand
    • What factors lead to CONSISTENT adoption?
    • Disease conditions
    • Patients’ disease progression
    • Drug action mechanism
    • Physician decision-making process
    • Drug sales representatives’ activities
  • Future Research
    • Random Utility Model of Prescribing Behavior?
    • Spillover effects
    • Opinion Leaders
    • Ethnolinguistic differences
    • Celebrex study: when do physicians reject new drugs?

◄Chapter 2► Do new drugs lead to better health outcomes?

Research Question

  • Do new drugs lead to better health outcomes?
  • More specifically, do patients who take Lipitor, Avandia, or Plavix experience a reduction in
    • ER visits, hospital admissions, hospital lengths of stay (problem?), and/or medical expenditures (compared to patients taking older drugs)?

Quote

  • “Too often,” says Robert Seidman, chief pharmacy officer at health insurer WellPoint, “we're choosing the newer, pricier drug without considering whether older drugs would get the job done just as well”
  • Lipitor: $612/180 20mg tablets
  • Zocor: $799/180 20mg tablets  but soon generics
  • Mevacor: $228.31/180 20mg tablets
    • www.drugstore.com prices

Literature Review

  • Lichtenberg (1996)
    • Number of hospital bed-days declined most rapidly for those diagnoses with the greatest change in the total number of drugs prescribed and greatest change in the distribution of drugs (proxy for novelty)
  • Lichtenberg (2001)
    • Patients who consume newer drugs experience fewer work-loss days than patients who consume older drugs; and the former tend to have lower non-drug expenditures, reducing total expenditures
  • Lichtenberg (2002)
    • With larger dataset, and 3 years instead of 1 year of observation, Lichtenberg argues that a reduction in the age of drugs decreased non-drug expenditures 7.2 times as much as it increased drug expenditures. (8.3 times for Medicare population)
  • Lichtenberg (2005)
    • Effect of the launch of new drugs: Average 1 week increase in life expectancy in the entire population

Conceptual Framework

  • Empirical question: Estimation of Average Treatment Effect
    • Are the high cost of new drugs justified based on their health outcome impact?
    • Lichtenberg studies do not address selection bias in treatment

Atorvastatin (Lipitor): Clinical Research

  • Collaborative Atorvastatin Diabetes Study (CARDS),
    • 2,800 patients with type-2 diabetes, no history of heart disease, and relatively-low levels of cholesterol,
  • Positive Health outcome:
    • patients who took Lipitor had a 37 percent reduction in major cardiovascular events
      • which included heart attacks, stroke, chest pain that required hospitalization, cardiac resuscitation, and coronary revascularization procedures.
    • 48 percent fewer Lipitor treated patients experienced strokes compared to those who received placebo
    • overall mortality rate for Lipitor patients was 27 percent lower than for those on placebo.
  • But: Study Sponsored by Pfizer / No comparison with older drugs / Relatively Healthy Population

Atorvastatin (Lipitor) Clinical Research - Hypertension

  • LIPITOR significantly reduced the rate of coronary events
    • either fatal coronary heart disease (46 events in the placebo group vs 40 events in the LIPITOR group)
    • or nonfatal MI (108 events in the placebo group vs 60 events in the LIPITOR group)]
    • relative risk reduction of 36% (based on incidences of 1.9% for LIPITOR vs 3.0% for placebo), p=0.0005
    • The risk reduction was consistent regardless of age, smoking status, obesity or presence of renal dysfunction. The effect of LIPITOR was seen regardless of baseline LDL levels. Due to the small number of events, results for women were inconclusive.
    • N = 10,305 (Anglo-Scandinavian Cardiac Outcomes Trial)
  • Source: www.lipitor.com

Mixed Results for Lipitor Vs. Zocor By THERESA AGOVINO, AP Business Writer Tuesday, November 15, 2005 06 57 PM

  • High doses of the cholesterol-lowering drug Lipitor were no better at preventing major heart problems than regular doses of rival Zocor, according to the latest study on efforts to aggressively treat the conditions released Tuesday.
  • Lipitor outperformed Zocor on several fronts such as lowering cholesterol and preventing nonfatal heart attacks. The findings will continue to give it an advantage in the market even if generic Zocor is less expensive, some doctors said.
  • But: HIGH DOSE OF LIPITOR vs. REGULAR DOSE OF ZOCOR
  • What about LIPITOR vs. MEVACOR, PRAVACHOL, LESCOL, CRESTOR

Empirical Strategy

  • Naïve Fixed Effects Regression

Threats to Identification

  • Selection for treatment most likely not random
  • Selection Bias in Treatment
    • Perhaps physicians assign nonrandom populations to treatment
    • Perhaps patients seek physicians who prescribe new drugs (e.g., Lipitor)

Correction for Selection Bias

  • Instrumental Variable Approach
    • Gives internally valid causal effects for individuals whose treatment status is manipulable by the instrument
      • Candidates: the combination of covariates from Chapter 2 as an instrument for the treatment (i.e., use of new drug, such as Lipitor)
      • With patient’s pre-adoption status in the instruments to avoid patient self-selection
      • However, may reduce statistical power
      • Note: we can see if patients actually self-select into treatment
    • But: instruments (predicts adoption) may also affect the dependent variable (measures for health outcome)?

Correction for Selection Bias

  • Selection on Observables
    • Propensity Score Matching
    • Analysis of the Effects of Unobservables?

Cost Analysis

  • Lipitor Costs (Taiwan NHID formulary 2004, in USD):
    • $1.04 per 10 mg tablet; $1.40 per 20 mg; $1.75 per 40 mg
  • What are the cost savings?
    • If new drug reduces emergency and hospital services
    • Savings = reduced cost in emergency and hospital services – increased drug costs
  • What are the additional costs?
    • If new drug has not health outcome impact?
    • Additional cost = difference in price of new and old drugs

Distribution of new Lipitor takers

“Treatment” vs. “Non-Treatment”

Graphical Evidence – ER visits No adjustment for selection bias

Graphical Evidence – ER visits (1009 Lipitor takers)

Graphical Evidence – Smoothed ER visits (1009 Lipitor takers)

Graphical Evidence – ER visits (656 consistent takers)

Graphical Evidence – Smoothed ER visits (656 consistent takers)

Graphical Evidence - Hospitalization

Graphical Evidence – Hospitalization (1009 Lipitor takers)

Graphical Evidence – Smoothed Hospitalization (1009 takers)

Graphical Evidence – Hospitalization(656 consistent takers)

Graphical Evidence – Smoothed Hospitalization(656 consistent takers)

Graphical Evidence – Average Length of Stay

Graphical Evidence – Average Length of Stay (1009 Lipitor takers)

Graphical Evidence – Smoothed average lengths of stay (1009 Lipitor takers)

Graphical Evidence – Average lengths of stay (656 consistent takers)

Graphical Evidence – Smoothed average lengths of stay (656 consistent takers)

Graphical Evidence – Average Hospital Expenditures

Graphical Evidence – Average expenditures (1009 takers)

Graphical Evidence – Smoothed average expenditures (1009 takers)

Graphical Evidence – Average expenditures (656 consistent takers)

Graphical Evidence – Smoothed average expenditures (656 consistent takers)

Discussion

  • Consistent Lipitor use does lead to better health outcomes?
    • Not just selection bias if consistency does improve health outcome?
    • But suggestive evidence that healthier patients are more likely to receive Lipitor consistently?
    • But: numerous possibilities for errors while merging
    • 2004 data consistently slightly “bizarre”
    • Treatment indicator very rough. 1 prescription of Lipitor over 2 visits = 50%  treatment
    • Need to consider consistency over a prescribed period of time: 3 months?
    • What did they take before Lipitor?
  • Need to include all indications for use of Lipitor
  • Need to adjust for patient heterogeneity
  • SELECTION ISSUES

Future Agenda

  • Better understand
    • Drug adoption decisions, based on chapter 1
    • Quest for proper instrument
    • Hypertension, Hypertensive Heart Diseases, Diabetes, High Cholesterol
    • Clinical trial results for the new drugs

Thank you for your attention and valuable assistance

◄Chapter 3► Do physicians consider patient out-of-pocket expenses when prescribing drugs?

Research Question

  • Do physicians consider patient out-of-pocket expenses?
    • In August 1999, Taiwan implemented a modest, linear prescription drug copayment system
    • Patients with one of 97 chronic conditions can be exempt from outpatient prescription copayments if physicians give an “chronic illness extended prescription certificate”
    • As of 2005, only 13% of eligible patient-visits receive the extended prescription certificate
  • Do physicians with high practice volume only give the extended prescription certificate? Or do patients have to demand the certificate?

Literature Review

  • Three bodies of literature:
    • Impact of cost-sharing on patients’ drug utilization choice:
      • Soumerai et al (1987, 1991, 1994), Nelson (1984), Tamblyn (2001).
    • Patient-Physician Principal-Agent Relationship
      • Especially: Supplier-induced demand (SID): Rice (1983), Yip (1998)
    • Physician consideration of patient out-of-pocket expenses
      • Only survey studies available:

Contribution and Limitation

  • Contribution
    • As far as I know, first paper to investigate through non-survey data whether physicians consider patient out-of-pocket expenses in their prescribing behavior
    • Policy implications:
  • Limitation
    • Copayment is insignificant (capped at $3.33 USD until 2001, then capped at $6.66 USD)
    • Generalizability?
    • Correlation Study

Conceptual Framework

  • Physicians generally earn greater income through increased practice volume
    • Physicians give certificates if the already have high practice volume
    • Or patients may demand certificate: proxied by competition and patient sophistication
    • Or both

Empirical Strategy

  • First: Fixed Effects Regression
    • Investigate effects of copayment on number of drugs, prescription duration, adjusted drug amount, and adjusted drug quantity

Empirical Strategy

  • Second: Logit/Probit Estimation
    • Effects of physician practice volume, patient sophistication level, and market competition on the likelihood of giving “extended prescription certificate”

Data Files

  • Ambulatory Care Expenditures by Visit
  • Details of Ambulatory Care Orders
  • Inpatient Expenditures by Admission
  • Details of Inpatient Orders
  • Expenditures for Prescriptions Dispensed at Contracted Pharmacies
  • Details of Prescriptions Dispensed at Contracted Pharmacies

Ambulatory Care Expenditures by Visit

Details of Ambulatory Care Orders

Inpatient Files

  • Inpatient Expenditures by Admission
    • Identification information; patient age and gender; date of admission and release; ICD9CM codes, ICD operation codes, DRG code, various fees, various copay amounts
  • Details of Inpatient Orders
    • Identification information; drug dispensed or services rendered

Summary Statistics, Master File



The database is protected by copyright ©sckool.org 2016
send message

    Main page