Journal of Mood & Anxiety Disorders 8 (2024) 100080

Contents lists available at ScienceDirect

Journal of Mood & Anxiety Disorders

journal homepage: www.journals.elsevier.com/journal-of-mood-and-anxiety-disorders

Association between prescriber practices and major depression

treatment outcomes
Sarah Rathnam a , Abhishek Sharma a , Kamber L. Hart b, c , Pilar F. Verhaak b, c ,
Thomas H. McCoy b, c , Roy H. Perlis b, c, * , Finale Doshi-Velez a, **
a
Harvard John A. Paulson School of Engineering and Applied Science, Cambridge, MA, United States
b
Center for Quantitative Health, Massachusetts General Hospital, Boston, MA, United States
c
Department of Psychiatry, Harvard Medical School, Boston, MA, United States

A R T I C L E I N F O A B S T R A C T

Keywords: Practice variability may represent an opportunity to improve care by identifying the differences in outcomes
Major depression associated with differences in practice. To characterize differences in depression treatment outcomes among
Primary care individual providers in outpatient psychiatry practices and primary care practices, we examined a longitudinal
Psychopharmacology
cohort derived from outpatient electronic health records from two academic medical centers and six community
Electronic health records
hospitals in Eastern Massachusetts. This cohort included antidepressant-treated individuals with an ICD-9/10
Health services
Antidepressants diagnosis of major depressive disorder, and deidentified health care providers treating at least 10 such pa-
tients per year between 2008 and 2022. We examined the association between individual provider prescribing
characteristics and proportions of treated patients who do not follow up after initial antidepressant prescription
or who achieve a stable ongoing prescription. In binomial regression models, among 104 psychiatrists, greater
heterogeneity in antidepressant prescribing and lesser proportion of serotonin reuptake inhibitors (SSRIs)1
prescribed were associated with greater rates of achieving stability (for heterogeneity, adjusted odds ratio AOR,
1.55 [95 % CI, 1.22 – 2.06]; for proportion of SSRIs, AOR, 0.01 [95 % CI, 0.00–0.59]). Among 369 primary care
physicians, greater volume of depression encounters per year, but not prescribing heterogeneity, was associated
with greater rates of achieving stability (for encounters, AOR, 2.15 [95 % CI, 1.61 – 2.89]; for heterogeneity,
AOR, 0.99 [95 % CI, 0.85 – 1.15]). Primary care and psychiatry predictors are not the same and therefore suggest
potentially distinct strategies to improve clinical outcomes in each setting. Trial Registration: N/A

1. Introduction typically ascribed to individual level differences, prompting efforts to

identify clinical subtypes [4] or predictive biomarkers [5,6]. However,
Clinical practice variability has emerged as an opportunity to iden- at least some of this variability may well arise from differences in
tify strategies to improve outcomes that does not require the develop- clinician practices.
ment of novel therapeutic or diagnostic tools [1,2]. Instead, Within psychiatry, such practice variation has rarely been studied.
understanding whether and why differences in practice associate with An analysis of commercial claims data found that individuals diagnosed
differential outcomes may allow identification of potential strategies with depression by psychiatrists were more likely to receive treatment of
that practices could adopt to improve clinical outcomes. any kind, and treatment perceived as adequate, than individuals diag-
Depression treatment outcomes are known to exhibit substantial nosed by primary care physicians [7]. Similar results emerged from
heterogeneity. A widely cited rule of thumb notes that 1/3 of individuals National Ambulatory Medical Care Surveys data comparing 1300 pri-
improve with initial treatment, and another 1/3 with multiple next-step mary care depression visits to 2418 psychiatry depression visits [8], as
treatments [3]. This variability in response to standard treatments is well as a survey comparing psychiatrists to primary care providers in

Abbreviations: SSRI, Selective serotonin reuptake inhibitor; MDD, Major depressive disorder; AD, Antidepressant; SNRI, Selective norepinephrine reuptake in-
hibitor; TCA, Tricyclic antidepressant; MCO, Managed Care Organization.
* Correspondence to: Center for Quantitative Health - Massachusetts General Hospital, 185 Cambridge Street, 6th Floor, Boston, MA 02114, United States.
** Correspondence to: Harvard John A Paulson School of Engineering and Applied Sciences, 29 Oxford Street, Cambridge, MA 02138, United States.
E-mail addresses: [email protected] (R.H. Perlis), [email protected] (F. Doshi-Velez).

https://doi.org/10.1016/j.xjmad.2024.100080
Received 20 May 2024; Received in revised form 6 August 2024; Accepted 6 August 2024
Available online 8 August 2024
2950-0044/© 2024 The Author(s). Published by Elsevier Inc. on behalf of Anxiety and Depression Association of America. This is an open access article under the
CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
S. Rathnam et al. Journal of Mood and Anxiety Disorders 8 (2024) 100080

their approach to treatment of a hypothetical mild major depressive captures continuation of a given AD in the 180-day period after index
episode [9]. While these studies compared primary and specialty care, prescription, with at least one recorded prescription in each 90d period.
they did not examine variability within psychiatric or primary care An index prescription was defined as the first AD prescription within the
practices, nor associations with outcomes. MGB health system, regardless of prior antidepressant treatment or
In the present study, we examined two large groups of clinicians subsequent prescription of a new AD. Only the initial AD prescription
spanning two health systems – one drawn from general outpatient within the MGB healthcare system was included in the data analysis to
psychiatry, the other from primary care psychiatry. We aimed to un- ensure that no patient was counted more than once.
derstand the extent to which outcomes inferred from electronic health Dropout reflects absence of any further AD prescriptions or other
records [10,11] varied among individual providers in these groups, and psychiatric treatment in the 90-day window after index prescription. We
then to quantify the extent to which such variability was explained by excluded patients completely lost to follow-up by excluding those with
quantifiable features of individual clinician practices. We hypothesized no facts in the electronic health record during the subsequent 90-day
that clinicians with greater clinical volume, and greater range of pre- period. Outcomes were attributed to the provider recording the initial
scribing to facilitate treatment personalization, would on average have AD prescription.
greater treatment stability and fewer treatment discontinuations among
their patients. 2.3. Analysis

2. Materials and methods We used standard univariate measures to describe distributions

among the providers in each of the 2 cohorts – outpatient psychiatry and
2.1. Study design and cohort development primary care. To examine the association between provider-level char-
acteristics and the two outcomes (proportion of individuals with stable
We included outpatient visits from the electronic health records of 2 treatment, and proportion with dropout), we used binomial regression –
academic medical centers and 6 community hospitals, and their affili- first univariate, then adjusted for all features of prescribing and clinical
ated sites, in Eastern Massachusetts between March 1, 2008, the start of volume. The threshold for statistical significance was considered to be 2-
routine e-prescribing across the hospital systems, and April 27, 2022. tailed p of 0.05. As a hypothesis-generating experiment – i.e., to identify
Each provider was assigned a unique anonymized identifier, linked to all possible differences for further study – we elected a priori not to correct
ICD-9 and 10 diagnostic codes, procedure codes, and prescriptions for testing multiple hypotheses.
associated with that provider.
We restricted the provider cohort to those billing for at least 10 in- 3. Results
dividuals with major depressive disorder (MDD) defined by diagnostic
codes (Supplemental Table 1) per year, averaged over the time period in The outpatient psychiatry provider cohort include 104 providers
which the provider appears in the data set, and at least 100 diagnostic who treated a total of 145,140 patients (see CONSORT diagrams, Sup-
codes billed in total. To account for the transition from ICD-9 to ICD-10, plemental Figures 1 and 2). On average, these providers treated 201
and reduce dimensionality of diagnostic codes, we aggregated codes up individuals per year (Table 1). In binomial regression examining asso-
to the 530 clinical categories defined by Clinical Classifications Software ciation between psychiatric provider practice characteristics and out-
Refined (CCSR, version 2022–1) [12]. As in prior work, we then comes (Fig. 1 and Supplemental Figure 3), we found that greater
generated high-dimensional representations for each provider based on heterogeneity in prescribing was associated with greater rates of sta-
counts of CCSR codes, reduced to two dimensions via UMAP [13,14] and bility (AOR, 1.55 [95% CI, 1.22 – 2.06]), as was lesser proportion of SSRI
then clustered into ten groups using a Gaussian Mixture Model [15]. prescriptions (AOR, 0.01 [95% CI, 0.00 – 0.59]). Lower volume of an-
This study protocol was approved by the appropriate institutional tidepressant prescribing per year was also significantly associated with
committee under protocol #2007P002138 with a waiver of informed greater stability (AOR, 0.35 [95% CI, 0.10 – 0.80]). However, not all
consent, as no contact with human subjects was required, data were patients met inclusion criteria for stability or dropout and were thus
deidentified, and consent would not be feasible at this scale. excluded from outcome analyses, resulting in a difference between the
providers included in the study cohort and those included in analyses for
2.2. Definitions of provider features and patient outcomes specific treatment outcomes.
We also examined regression models for proportion of dropout
For each provider, we calculated the total number of antidepressant (Fig. 2 and Supplemental Figure 4), using the same features. In these
(AD) prescriptions recorded, then divided by the provider’s number of models, lower volume of antidepressant prescribing was associated with
total diagnostic codes as a proxy for practice volume. We also examined lesser rates of drop out (AOR, 0.66 [95% CI, 0.42 – 0.98]). No other
proportion of total prescriptions representing SSRI, selective norepi- prescribing or volume characteristics were significantly associated with
nephrine reuptake inhibitor (SNRI), atypical antidepressant, or tricyclic early discontinuation. We excluded patients completely lost to follow-up
antidepressant (TCA). Because augmenting agents are not routinely used by excluding those with no facts in the electronic health record during
in primary care settings, we chose to focus solely on primary antide-
pressants. We generated a measure of heterogeneity of medications used Table 1
by each provider by determining the number of different ADs accounting Characteristics of prescribing clinicians.
for 75% of all AD prescriptions for each provider. In essence, this rep- Primary Care Providers Psych Providers
resents the breadth or range of prescribing for a given provider.
Count (%) 369 (78.01%) 104 (21.99%)
Treatment outcomes were defined for all individuals with at least one Mean (SD) Mean (SD)
diagnostic code of MDD, excluding those with codes for schizophrenia or Antidepressant heterogeneity 2.19 (1.17) 2.90 (1.93)
bipolar disorder. As in prior work [11], we examined treatment stability, Antidepressants/year 8.61 (9.25) 15.11 (36.83)
reflecting the proportion of patients who remain on a stable AD treat- MDD patients/year 106.67 (175.46) 57.49 (61.44)
MDD encounters/year 50.40 (83.30) 106.43 (199.25)
ment, and treatment dropout, reflecting the proportion of patients who
Psychiatry encounters/year 129.60 (207.43) 194.99 (287.59)
do not return after an initial prescription. Although clinical trial out- Total encounters/year 610.76 (761.69) 201.35 (289.95)
comes such as response and remission are challenging to reliably define % SSRI 58.74 (35.49) 46.74 (35.29)
for individuals using solely coded clinical data, we applied a simplifying % SNRI 15.66 (26.85) 22.39 (30.06)
but face-valid assumption that successful treatments continue uninter- % TCA 8.48 (19.76) 7.08 (18.75)
% Other antidepressant 9.06 (19.68) 10.29 (24.30)
rupted over time with repeated prescriptions. Specifically, stability

2
S. Rathnam et al. Journal of Mood and Anxiety Disorders 8 (2024) 100080

Fig. 1. Adjusted binomial regression model analyzing psychiatric provider practice characteristics associated with treatment stability.

Fig. 2. Adjusted binomial regression model analyzing psychiatric provider practice characteristics associated with treatment dropout.

the subsequent 90-day period. individuals per year (Table 1). In a regression model for treatment sta-
We next sought to understand whether similar patterns emerged in bility, greater number of depression encounters was associated with
the primary care provider cohort. This group included 369 providers greater rates of depression stability (AOR, 2.15 [95% CI, 1.61 – 2.89]),
who treated a total of 268,018 patients. (see CONSORT diagrams, but no significant association with prescribing heterogeneity was
Supplemental Figures 1 and 2). On average, these providers treated 611 detected in adjusted regression models (AOR, 0.99 [95% CI, 0.85 –

Fig. 3. Adjusted binomial regression model analyzing primary care providers practice characteristics associated with treatment stability.

3
S. Rathnam et al. Journal of Mood and Anxiety Disorders 8 (2024) 100080

1.15]); (Fig. 3 and Supplemental Figure 5). treatments, and thus providers who used their full range of options to
For dropout, fewer depression encounters were associated with identify treatments got better results. While the underlying mechanism
lower dropout rates (AOR, 0.36 [95% CI, 0.30 – 0.42]); (Fig. 4 and of association merits further study, our work suggests that there may be
Supplemental Figure 6), as were lesser proportion of SSRI, tricyclic, and an opportunity for educational interventions promoting comfort with a
other antidepressant prescribing (for SSRI, AOR, 0.68 [95% CI, 0.48 – broader range of prescribing to facilitate greater rates of achieving sta-
0.95]; for tricyclic, AOR, 0.42 [95% CI, 0.27 – 0.66]; for other antide- ble treatment. (If prescribing heterogoneity were solely a proxy for
pressants, AOR, 0.57 [95% CI, 0.37 – 0.89]). Conversely, greater het- treating more severely ill individuals requiring more complex treatment
erogeneity of prescribing was associated with greater rates of regimens, we might have expected the opposite finding, with lower
discontinuation (AOR, 1.12 [95% CI, 1.04 – 1.21]). stability associated with greater heterogeneity). Conversely, among
primary care clinicians, outcomes were more positive among providers
4. Discussion with greater volume of major depressions visits – both in terms of
achieving stability and minimizing dropout. Greater heterogeneity was
In these two cohorts of providers treating major depressive disorder, actually associated with higher dropout rates. While these associations
we identified distinct correlates of achieving stable antidepressant cannot establish causation, this finding may suggest the value of aug-
treatment. Notably, heterogeneity of antidepressant prescribing was menting training in depression treatment among primary care phys-
associated with greater likelihood of stability only among psychiatric icicans, or interventions that could increase visit frequency in this
providers; among primary care providers, this feature exhibits no asso- setting. That is, primary care physicians who see individuals with
ciation with stability, and indeed was associated with greater rates of majore depression more frequently may achieve better results, con-
dropout. Conversely, volume of visits for major depression had no sig- sisitent with quality guidelines for follow-up that may not be fully
nificant association with outcome among psychiatric providers, while it implemented [18,19]. On the other hand, in this group of clinicians, in
associated with greater likelihood of stability, and lesser likelihood of contrast to psychiatry, treatment heterogeneity was not significantly
dropout, among primary care providers. We hypothesize that a greater associated with stability in adjusted models. This result suggests that
range of prescribing among individuals with less psychopharmacologic simply broadening pharmacologic options for primary care physicians
training may lead to greater discontinuation because of challenges in (rather than, for example, focusing on strategies to improve adherence
managing side effects or adjusting dose. Future work will be needed to for first-line treatments, or facilitating referral for more difficult-to-treat
examine this possibility directly. patients) may not be a useful strategy.
Our work is difficult to compare directly to prior studies, which have This study has multiple limitations. While it includes a broad group
generally sought to compare clinical populations rather than capture of clinics across two health systems, the extent to which this region will
variability within them. Those analyses generally found greater rates of generalize to others in the US and internationally remains to be deter-
antidepressant prescribing among psychiatrists compared to primary mined. Furthermore, it does not include the more sensitive outcome
care physicians [16]. We did not directly compare these two groups, measures typically available in clinical trials; on the other hand, we
except for descriptive purposes, reasoning that the patients treated in elected to focus on measures which are face-valid in terms of relevance
these settings are likely to be quite different. For example, individuals to clinicians and patients, and available in any electronic health record.
with more severe depression, or greater comorbidity, would be likely to These measures were also demonstrated to be modestly predictable on
be referred for specialty care rather than remaining in a primary care the basis of patient-level data in prior work [8,9], suggesting that
setting. One prior investigation compared providers from different kinds provider-level data would also be informative. While our investigation
of practices, contrasting staff/group-model Managed Care Organizations aimed to examine differences among providers, certain aspects of
(MCOs) to network-model organizations; this study found differences in provider-level data, including level of training, were inaccessible due to
propensity to refer out to specialists, rather than treating without IRB limitations. Furthermore, patient characteristics that were not
referral [17]. analyzed, such as insurance status, could have influenced physician type
Our work may suggest opportunities to improve the standard of care as well as drop out and treatment outcomes. Results were also not
for depression treatment and indicates that distinct strategies may be adjusted for multiple comparisons using a Bonferroni-corrected
prioritized in different settings. Among outpatient psychiatrists, one threshold and thus observed associations should be treated as hypoth-
conjecture for why greater prescribing variability was associated with eses warranting further investigation.
better outcomes is that patients in specialist care need more specific Our work may suggest opportunities to improve the standard of care

Fig. 4. Adjusted binomial regression model analyzing primary care providers practice characteristics associated with treatment dropout.

4
S. Rathnam et al. Journal of Mood and Anxiety Disorders 8 (2024) 100080

for depression treatment and indicates that distinct strategies may be relationship with Mila Health that includes: consulting or advisory. Roy
prioritized in different settings. Among outpatient psychiatrists, one H. Perlis reports a relationship with Alkermes Inc that includes:
conjecture for why greater prescribing variability was associated with consulting or advisory. Roy H. Perlis reports a relationship with
better outcomes is that patients in specialist care need more specific Genomind that includes: consulting or advisory. Roy H. Perlis reports a
treatments, and thus providers who used their full range of options to relationship with Takeda Pharmaceuticals America Inc that includes:
identify treatments got better results. While the underlying mechanism consulting or advisory. Roy H. Perlis reports a relationship with Circular
of association merits further study, our work suggests that there may be Genomics that includes: consulting or advisory and equity or stocks. Roy
an opportunity for educational interventions promoting comfort with a H. Perlis reports a relationship with Psy Therapeutics that includes:
broader range of prescribing to facilitate greater rates of achieving sta- consulting or advisory and equity or stocks. JAMA Network - Open
ble treatment. (If prescribing heterogeneity were solely a proxy for (Service as an Associate Editor) RHP If there are other authors, they
treating more severely ill individuals requiring more complex treatment declare that they have no known competing financial interests or per-
regimens, we might have expected the opposite finding, with lower sonal relationships that could have appeared to influence the work re-
stability associated with greater heterogeneity). Conversely, among ported in this paper.
primary care clinicians, outcomes were more positive among providers
with greater volume of major depression visits – both in terms of Acknowledgements
achieving stability and minimizing dropout. Greater heterogeneity was
actually associated with higher dropout rates. While these associations The authors report not have any acknowledgements to report.
cannot establish causation, this finding may suggest the value of aug-
menting training in depression treatment among primary care physi- Appendix A. Supporting information
cians, or interventions that could increase visit frequency in this setting.
That is, primary care physicians who see individuals with major Supplementary data associated with this article can be found in the
depression more frequently may achieve better results, consistent with online version at doi:10.1016/j.xjmad.2024.100080.
quality guidelines for follow-up that may not be fully implemented [16,
17]. On the other hand, in this group of clinicians, in contrast to psy- References
chiatry, treatment heterogeneity was not significantly associated with
stability in adjusted models. This result suggests that simply broadening [1] Galanter W, Eguale T, Gellad W, Lambert B, Mirica M, Cashy J, et al. Personal
formularies of primary care physicians across 4 health care systems (DOI:) JAMA
pharmacologic options for primary care physicians (rather than, for Netw Open 2021;4(7):e2117038. https://doi.org/10.1001/
example, focusing on strategies to improve adherence for first-line jamanetworkopen.2021.17038.
treatments, or facilitating referral for more difficult-to-treat patients) [2] Cordoba G, Siersma V, Lopez-Valcarcel B, Bjerrum L, Llor C, Aabenhus R, et al.
Prescribing style and variation in antibiotic prescriptions for sore throat: cross-
may not be a useful strategy. However, specific care location could sectional study across six countries (DOI:) BMC Fam Pr 2015;16:7. https://doi.org/
associate with outcome differences. Because there is not 1:1 relationship 10.1186/s12875-015-0224-y.
between providers and locations, or between patients and locations, [3] Rush AJ, Trivedi MH, Wisniewski SR, Nierenberg AA, Stewart JW, Warden D, et al.
Acute and longer-term outcomes in depressed outpatients requiring one or several
providers in these networks may work in multiple clinics and patients treatment steps: a STAR*D report. Am J Psychiatry 2006;163(11):1905–17.
may interact with the systems in multiple settings (e.g., primary care, [4] van Loo HM, de Jonge P, Romeijn JW, Kessler RC, Schoevers RA. Data-driven
cardiology, outpatient psychiatry). This is why we elected to focus on subtypes of major depressive disorder: a systematic review. BMC Med 2012;10:
156.
individual provider-level characteristics, rathern than properties of an
[5] Perlis RH. Pharmacogenomic testing and personalized treatment of depression.
individual clinic or setting, reasoning that these characteristics may be Clin Chem 2014;60(1):53–9.
more straightforward to identify and modify. However, future work [6] Schmidt HD, Shelton RC, Duman RS. Functional biomarkers of depression:
should further investigate the association between properties of indi- diagnosis, treatment, and pathophysiology (DOI:) Neuropsychopharmacology
2011;36(12):2375–94. https://doi.org/10.1038/npp.2011.151.
vidual settings and differences in patient outcomes. [7] Kniesner TJ, Powers RH, Croghan TW. Provider type and depression treatment
Overall, we identified associations between variations in practice adequacy (DOI:) Health Policy 2005;72(3):321–32. https://doi.org/10.1016/j.
and clinical outcomes which differ for two groups of providers. Future healthpol.2004.09.008.
[8] Stafford RS, Ausiello JC, Misra B, Saglam D. National patterns of depression
work should examine whether altering these practice characteristics can treatment in primary care. Prim Care Companion J Clin Psychiatry 2000;2(6):
improve outcomes. As all of our measures and outcomes are readily 211–6.
derived from electronic health records, investigating potential in- [9] Lawrence RE, Rasinski KA, Yoon JD, Meador KG, Koenig HG, Curlin FA. Primary
care physicians’ and psychiatrists’ approaches to treating mild depression (DOI:)
terventions either naturalistically or with randomized designs should be Acta Psychiatr Scand 2012;126(5):385–92. https://doi.org/10.1111/j.1600-
highly feasible. 0447.2012.01887.x.
[10] Pradier MF, McCoy TH, Hughes M, Perlis RH, Doshi-Velez F. Predicting treatment
dropout after antidepressant initiation. Transl Psychiatry 2020;10(1):60.
Funding [11] Hughes MC, Pradier MF, Ross AS, McCoy TH, Perlis RH, Doshi-Velez F. Assessment
of a prediction model for antidepressant treatment stability using supervised topic
This work was supported by the National Institutes of Health models. JAMA Netw Open 2020;3(5):e205308.
[12] Clinical Classifications Software Refined (CCSR) [Internet]. [cited 2023 Dec 4].
[R01MH123804-03; Dr. Perlis and Dr. Doshi-Velez]. The sponsors did
[13] McInnes L., Healy J., Melville J. UMAP: Uniform Manifold Approximation and
not contribute to any aspect of study design, data collection, data Projection for Dimension Reduction [Internet]. arXiv; 2020 [cited 2023 Dec 4].
analysis or data interpretation. The authors had the final responsibility DOI: 10.48550/arXiv.1802.03426.
for the decision to submit for publication. [14] Maaten L, van der, Hinton G. Visualizing data using t-SNE. J Mach Learn Res 2008;
9(86):2579–605.
[15] McLachlan GJ, Rathnayake S. On the number of components in a Gaussian mixture
Declaration of Competing Interest model. Wiley Int Rev Data Min Knowl Disc 2014;4(5):341–55.
[16] Pincus HA, Tanielian TL, Marcus SC, Olfson M, Zarin DA, Thompson J, et al.
Prescribing trends in psychotropic medicationsprimary care, psychiatry, and other
The authors declare the following financial interests/personal re- medical specialties (DOI:) JAMA 1998;279(7):526–31. https://doi.org/10.1001/
lationships which may be considered as potential competing interests: jama.279.7.526.
Roy H. Perlis reports financial support was provided by National In- [17] Meredith LS, Rubenstein LV, Rost K, Ford DE, Gordon N, Nutting P, et al. Treating
depression in staff-model versus network-model managed care organizations (DOI:
stitutes of Health. Roy H. Perlis reports a relationship with Vault Health
that includes: consulting or advisory and equity or stocks. Roy H. Perlis
reports a relationship with Belle Artificial Intelligence that includes:
consulting or advisory. Roy H. Perlis reports a relationship with Swan AI
Studios that includes: consulting or advisory. Roy H. Perlis reports a

5
S. Rathnam et al. Journal of Mood and Anxiety Disorders 8 (2024) 100080

) J Gen Intern Med 1999;14(1):39–48. https://doi.org/10.1046/j.1525- [19] Simon GE, VonKorff M, Rutter C, Wagner E. Randomised trial of monitoring,
1497.1999.00279.x. feedback, and management of care by telephone to improve treatment of
[18] Lage I, Pradier MF, McCoy TH, Perlis RH, Doshi-Velez F. Do clinicians follow depression in primary care. BMJ 2000;320(7234):550–4.
heuristics in prescribing antidepressants? J Affect Disord 2022;311:110–4.