Running some tests: doctor-patient demographic concordance and diagnostic test orders 17/ 4/2014 Singapore Health Economics Association Conference Sophie Joyce Post-doctoral Research Fellow at SKBI, Singapore Management University Introduction • Study association between doctor-patient gender and ethnic group concordance and the amount of diagnostic tests (laboratory and radiology) ordered during a hospital stay • Find a statistically signiﬁcant reduction in laboratory and radiology test orders with demographic concordance. Over-use of diagnostic tests • Increased use of diagnostic tests: - American College of Physicians publishes lists of unnecessary tests and procedures (Cassel and Guest, 2012) ⁻ Costs of unnecessary diagnostic tests and procedures has been estimated at USD 200 and 250 billion in the U.S. (Berwick and Hackbarth, 2012; Thompson, 2011) • ⁻ ⁻ ⁻ Risk to patient health: exposes patients to harmful radiation false positive clinical abnormality - excessive medical treatment ‘Sifting through and assigning importance to vast quantities of unfiltered information is a defining skill of our [medical] profession. One of the unintended yet unavoidable consequences of increased diagnostic information is increased false positives.’ (Sutton, 2011, pg.1600) Background Demographic concordance in doctor-patient consultation: Medical literature - improved communication and patient satisfaction in demographically concordant consultations (LaVeist and Nuru-Jeter, 2002; Street Jr et al., 2007, 2008; Sandhu et al., 2009, Van Ryn and Burke, 2000; Gordon et al., 2006). Economic literature - Godager (2012) patients prefer a GP of the same gender. Balsa and McGuire (2005) model poorer patient communication in racially discordant consultations to explain racial variation in medical treatments. Data • Data from a large metropolitan hospital in New Zealand • Study period is 2004 to 2011, approximately 236,205 inpatient events in regression analysis • Ethnic and gender information on doctors obtained from HR department – associate one of 7 ethnicity groupings to doctors (European, Māori, Pacific Peoples, Asian, Indian, Middle Eastern, and Latin American/other) • 324 doctors with information on both gender and ethnic group: - 82.4% are European ethnic origin, 7.4% are Asian ethnic origin, 4.9% are Indian ethnic origin - 59.3% are Male Empirical method • Specify a fixed-effect on an inpatient event i’s primary doctor, j: _ = + + 1 + 2 ℎ + 3 __ℎ + + • Demographic concordance is estimated with a set of dummy variables for each type of concordance; ethnic only, gender only and ethnic plus gender. The base category is no demographic concordance. • Dummy variables for patient’s gender and ethnic group control for variation in average severity or complexity of illnesses across patient demographic groups. Assignment of patients to doctors in hospital Are variables for (doctor-patient) gender and/or ethnic group concordance exogenous? • Use the primary doctor identified in hospital data to determine whether demographic concordance occurred. • Patients are assigned to doctors on-call at patient’s arrival time to hospital, no policy for demographic matching • Patient or doctor can request transfer to another doctor - therefore cannot guarantee random assignment of doctors to patients • Argue any ‘informal’ sorting of patients to doctors, for demographic reasons, would not be expected to explain my negative relationship: - Expect transfers, for demographic reasons, to achieve concordance rather than discordance > concordant patients treated by more doctors -> positive, rather than negative, bias in diagnostic test orders. • In addition, I am also able to restrict my sample to patients that have the same doctor recorded for their admittance, primary and discharge from hospital. This allows me to check that patients who transfer doctors, for whatever reason, do not explain my results. Demographic concordance variables • Mutually exclusive demographic concordance categories, base category is no demographic concordance. • - Of the patient population in study: 25.2% ethnic only concordance (e.g. European patient and doctor) 24.8% gender only concordance 31% ethnic plus gender concordance (e.g. European and male doctor and patient) 19% no demographic concordance • 97.2% of ethnic concordance is European, 57% of gender concordance is male • A large proportion of ethnic concordance is European, this is a limitation of my data. I further estimate the relationship between gender concordance and diagnostic test orders for male and female doctors separately. Laboratory resource outcomes • Total cost and quantity of laboratory tests Kdensity square root laboratory cost .08 .04 .06 Density .003 .002 0 0 .02 .001 Density .004 .1 .005 Kdensity raw laboratory cost 0 500 1000 Raw laboratory cost 1500 2000 0 99th percentile of raw lab. cost excluded Quantity of laboratory tests 20 30 square root laboratory cost 40 50 .2 0 .1 .05 .1 Density .3 .15 .4 .2 Kdensity logarithm laboratory cost 0 Density 10 99th percentile of square root lab. cost excluded 0 50 100 Laboratory quantity 99th percentile of lab. quantity excluded 150 200 0 2 4 Logarithm laboratory cost 99th percentile of logarithm lab. cost excluded 6 8 Radiology resource outcomes Total cost of radiology tests and a binary outcome for whether patient received one or more radiology test during hospital stay Kdensity square root radiology cost Density 0 0 .002 .05 .004 Density .1 .006 .15 .008 Kdensity raw radiology cost 0 1000 2000 Raw radiology cost 3000 4000 0 99th percentile of raw rad. cost excluded 20 40 square root radiology cost 60 99th percentile of square root rad. cost excluded .2 0 0 .1 .2 .4 Density .3 .6 .4 Quantity of radiology tests .8 Kdensity logarithm radiology cost Density • 2 4 6 Logarithm radiology cost 99th percentile of logarithm rad. cost excluded 8 10 0 5 Radiology quantity 99th percentile of rads. quantity excluded 10 15 Econometric model • Cost outcomes - Fixed-effect Ordinary Least Squares (OLS) regression on, raw, natural-logarithm transformed, square-root transformed costs • Quantity outcomes - Radiology – binary – fixed-effect linear - Laboratory – count – fixed-effect linear and poisson • 99th percentile cost and quantity excluded, standard errors clustered on doctor Hospital Study Sample • All hospital inpatient events during 2004-2011, excluding: - Entry to hospital from waiting list (i.e. for an elective procedure) - Aged under 5 years - Discharged from a rehabilitation facility Explanatory variables • Explanatory variables: Patient Characteristics: ⁻ Male patient ⁻ Patient age ⁻ Socioeconomic scale ⁻ Patient ethnic group dummy variables Timing of admission: ⁻ Day of week admission ⁻ Year of admission ⁻ Time of day of admission Type of admission: ⁻ Transfer patient ⁻ Accident and Emergency Department (AED) entry ⁻ Acute admission (relative to base category of arranged admission) ⁻ Intended day patient admission ⁻ Accident as cause of hospital admission Clinical condition: ⁻ Whether a patient had one or more inpatient events within the last 60 days ⁻ Length of hospital stay in days ⁻ The number of diagnoses ⁻ If patient had a surgical theatre event ⁻ Charlson’s comorbid conditions ⁻ Major Diagnostic Category of illness. Laboratory test: results Laboratory test: results Radiology test: results Robustness tests • • • Fixed-effect on doctor and Major Diagnostic Category (MDC) Admit, primary and discharge doctors are the same Including manually associated ethnic and gender groups for doctors Robustness tests • Specific medical populations – most populous Major Diagnostic Categories in hospital data (excluding pregnancy) Robustness tests • Male and female doctors in MDCs of 11 (kidney), 6 (digestive), 5 (circulatory) and 4 (respiratory) Explaining a negative relationship • Find no evidence that demographic concordance is positively associated with patient health outcomes of mortality within 60 days of hospital admission and emergency readmission to hospital Suggests patients in demographically concordant consultations received adequate health care. • Suggest improved information on patient’s health condition from consultation process itself (communication, physical exam) and/or preferences for fewer diagnostic tests could explain a reduction in diagnostic resources. • Health care publicly provided and doctors not at risk of medical malpractice lawsuits, therefore insurance arrangements or fear of litigation is not expected to explain a negative relationship. Conclusion • Statistically significant reduction in laboratory and radiology resources in demographically concordant pairs. Though, estimated size of reduction in diagnostic resources is small. • Gender concordance more robust than ethnic concordance (limited variation in ethnic groups of doctors.) • Reducing ‘unnecessary’ diagnostic test orders reduces complexity in decision-making and risk of over-treatment. Could occur through ‘improvements’ in doctor-patient relationship.