Federal Reserve Bank of Chicago

Physician Practice Style for Mental Health Conditions: The Case of ADHD

Kelli Marquardt

August 2021

WP 2022-22

https://doi.org/10.21033/wp-2022-22

*Working papers are not edited, and all opinions and errors are the responsibility of the author(s). The views expressed do not necessarily reflect the views of the Federal Reserve Bank of Chicago or the Federal Reserve System.

Physician Practice Style for Mental Health Conditions: The

Case of ADHD

Kelli Marquardt

Federal Reserve Bank of Chicago

The University of Arizona

kmarquardt@frbchi.org

August 2021

Abstract

There is a robust literature documenting the importance of physician practice style (e.g., the propensity to perform certain operations) in explaining outcomes related to patients' physical health. Yet little is known about the role of physicians in explaining patients' mental health outcomes. This paper uses novel data on doctor note text together with natural language processing techniques to estimate and document heterogeneity in physician practice style for diagnosing Attention Decit Hyperactivity Disorder (ADHD). I nd signicant variation in both diagnostic intensity (the mean propensity to diagnose) and diagnostic compliance (the weight that physicians place on medical guidelines). Physician characteristics can explain some of this heterogene- ity. Specically, both female physicians and recent graduates have higher diagnostic compliance and lower diagnostic intensity than their respective counterparts. Mental health diagnostic errors lead to excess medical and societal spending. Given the costs of such errors, the ndings in this paper encourage a re-evaluation of the mental health identication process, though perhaps targeted at specic sub-groups of physicians.

Keywords: Physician Practice Style, Child Mental Health, Textual Analysis.

JEL classication: I10, J24, C8.

  • This paper is based upon work supported by the University of Arizona Graduate and Professional Student Council, Research and Project (ReaP) Grant -2019. Data provided by the University of Arizona
    Center for Biomedical Informatics & Biostatistics, Department of Biomedical Informatics.
    The views here do not represent those of the Federal Reserve Bank of Chicago or the Federal Reserve System.

1

  • Introduction

Health care spending and patient outcomes vary dramatically both across and within markets (Kibria et al., 2013). Heterogeneity in physician practice style is a widely documented source of such variation. To reduce ineciency in health spending, both policy-makers and researchers are interested in estimating physician practices style and understanding why physicians treat patients di erently. The existing literature on this topic has typically used insurance claims or hospital discharge data to estimate physician practice style in physical health applications. For example, Epstein and Nicholson (2009) document heterogeneity across obstetricians in their propensity to preform C-sections. Currie et al. (2016) explore variation in the cardiologist's decision to preform invasive procedures for heart-attack pa- tients. Gowrisankaran et al. (2018) examine the correlation of physician practice style across three conditions: Angina, Appendicitis, and Transient Ischemic Attacks.

While the extant literature shows the importance of physician practice style in physical health applications, little is known about the inuence of the physician on patient mental health outcomes. In 2019, mental health sector spending was estimated at $225 billion, more than a 50% increase since 2009 (Open Minds, 2020). Early and accurate detection of mental health conditions could help reduce these cost burdens, both on an individual and national level. Therefore, it is important to understand the role of physicians in the mental health diagnostic decision making process. The goal of this paper is to quantify physician practice style as it relates to mental health diagnosis and to document heterogeneity across physicians as a potential source of variation in mental health care spending.

The challenge of quantifying practice style for mental health conditions stems from the process of diagnosis. The presence of a mental health condition cannot be determined via any blood test or medical imaging. Instead, a physician must conduct a behavioral interview and match subjective patient symptoms to those which dene a diagnosis, outlined in The Diagnostic and Statistical Manual of Mental Disorders, which is currently in its fth edition (DSM-V).Because these symptoms are expressed to the physician in a more conversational manner, they are not typically denoted in traditional health datasets, and therefore are not observable to the econometrician. In this paper, I propose using a new

2

source of data, namely doctor note text from electronic health records, to overcome the partial observability problem and quantify physician practice style for a speci c mental health condition, Attention De cit Hyperactivity Disorder (ADHD).

ADHD is an ideal application for this study for two reasons. First, ADHD is the most prevalent child mental health condition, being diagnosed in nearly 10% of children worldwide. It is a costly condition, impacting individual children, families, and society. Doshi et al. (2012) estimate the annual economic impact of ADHD diagnosis in the range of $143-$266 billion US dollars. Second, despite the documented large costs associated with ADHD diagnostic errors, recent research suggests that this condition is often inaccurately diagnosed in practice (Merten et al., 2017).1 Estimating the role that physicians play in the ADHD diagnosis decision can help inuence medical and health care policy aimed at reducing ADHD diagnostic errors and their associated costs.

I obtain electronic health record data from a large healthcare system in Arizona, which importantly includes access to de-identi ed clinical doctor notes for over 12,311 pediatric patients. I use these data to estimate and document heterogeneity in physician practice style for 129 unique physicians. I rst present a natural language processing (NLP) algorithm which I apply to clinical doctor note texts and derive how appropriate" a patient is for an ADHD diagnosis based on symptom match. The NLP algorithm takes the patient record as an input and produces a single output metric, which can be interpreted as how closely the patient's expressed symptoms overlap with the ADHD speci c symptoms de ned by the DSM. I then use this patient metric as a control in a diagnosis decision-making model and estimate physician speci c parameters. The intercept e ect, which I call diagnostic intensity, measures the mean propensity to diagnose. The slope e ect, which I call diagnostic compliance, measures how closely the physician follows national diagnostic guidelines when diagnosing a child with ADHD. Together, these two components de ne the physician practice style, providing a quantitative measure of mental health diagnosis quality.

1ADHD misdiagnosis has been both quantitatively and qualitatively examined across many disciplines. Within health economics, scholars have used birthdate to school entry cut-o date as a discontinuity predicting potential over diagnosis of ADHD (e.g., Elder, 2010; Schwandt and Wuppermann, 2016; Persson et al., 2021). Health services and psychology literature explore this topic using meta analyses (Sciutto and Eisenberg, 2007) and physician/patient surveys (Chan et al., 2005; Bruchmuller et al., 2012).

3

In an ideal setting, physicians would have sucient time and skill in their behavioral assessments, resulting in high diagnostic compliance and low diagnostic intensity. In other words, the physician would follow the DSM-V guidelines and diagnose a patient with ADHD if and only if they meet the necessary requirements for diagnosis. However, due to a variety of factors including time and information constraints, physicians do not diagnose in an ideal setting. I nd signicant variation in physician practice style for ADHD, implying that physicians choose dierent diagnosis codes even for patients with identical set of behavioral symptoms. The median physician in my sample has a diagnostic intensity of -0.17 and diagnostic compliance of 0.44. The negative intensity estimate suggests that the median physician is, on average, risk-averse when choosing to diagnose their patient with ADHD. The positive (but small) diagnostic compliance estimate suggests that physicians put some weight on ocial DSM-V guidelines, but also likely rely on prior beliefs and/or other patient signals when making the diagnosis decision.

I then use hand-collected information on physician background to estimate how much of the variation in physician practice style can be explained by physician characteristics and training. I nd that physician gender and experience are the strongest predictors of physician practice style for ADHD with both female physicians and recent graduates having higher diagnostic compliance and lower diagnostic intensity than their respective counterparts. Because deviation from medical guidelines can result in diagnostic errors and unnecessary spending, these results suggest a re-evaluation of how mental health conditions are identied with a focus on the heterogeneity in physician guideline adherence.

The rest of the paper is outlined as follows. In the next section, I describe the data and provide background medical information on the ADHD diagnostic process. Section 3 details the methodology needed for mental health practice style estimation. This includes a rst stage natural language processing algorithm and second stage diagnosis decision-making model. Section 4 presents the main results and heterogeneity analysis. Finally, Section 5 concludes with a discussion and extensions for future work.

4

Attachments

  • Original Link
  • Original Document
  • Permalink

Disclaimer

Altri SGPS SA published this content on 23 June 2022 and is solely responsible for the information contained therein. Distributed by Public, unedited and unaltered, on 23 June 2022 16:45:05 UTC.