I

Aprweo

aH

Women’s Health Issues 16 (2006) 122–132

CP

DOES PATIENT–PROVIDER GENDER CONCORDANCEAFFECT MENTAL HEALTH CARE RECEIVED BY PRIMARY

CARE PATIENTS WITH MAJOR DEPRESSION?

Kitty S. Chan, PhD,a* Chloe E. Bird, PhD,a Robert Weiss, PhD,b Naihua Duan, PhD,b,c

Lisa S. Meredith, PhD,a and Cathy D. Sherbourne, PhDa

aRAND Corporation, Health Program, Santa Monica, CaliforniabUCLA Department of Biostatistics, Los Angeles, California

cUCLA Department of Psychiatry and Biobehavioral Sciences, Los Angeles, California

Received 20 May 2005; revised 1 February 2006; accepted 24 February 2006

Purpose. We sought to determine whether patient–provider gender concordance influencesthe detection and care of depression and comorbid anxiety and substance use in patients withmajor depression

Methods. Cross-sectional analyses of baseline patient survey data linked with provider datawere performed. Data based on routine primary care visits in clinics from a variety of healthsystems serving diverse patient populations across the United States. Participants all hadmajor depression. Depression care was examined in the Quality Improvement for Depression(QID) Collaboration sample (n patients � 1,428, n providers � 389). In a subanalysis of datasolely from 714 patients and 157 providers from Partners-In-Care, one of the projectsparticipating in QID, we also examined detection of anxiety disorder and alcohol or drugproblems.

Main Findings. Rates of detection and care of mental health problems in primary care werelow even among patients with major depression. Except for anxiety counseling in femalepatients, patient–provider gender concordance did not improve care as hypothesized. How-ever, female providers were more likely to counsel on anxiety and less likely to counsel onalcohol or drug use than male providers. Female patients were less likely to be counseled onalcohol or drug use compared with male patients.

Conclusion. Detection and care of mental health and substance use problems for patients withmajor depression is not influenced by patient–provider gender concordance. However,depressed female patients may have greater unmet needs for alcohol and drug use counseling

than their male counterparts.

r(2(ead(amh

ntroduction and Background

ppropriate care for depression, which may includesychotherapy and medication use, has been shown toeduce the personal and societal burden for patientsith this condition (Schoenbaum et al., 2002). How-

ver, diagnosis and treatment rates for depression andther mental health and substance abuse problems

* Correspondence to: Kitty S. Chan, Department of Health Policynd Management, Johns Hopkins Bloomberg School of Publicealth, 624 North Broadway, Room 644, Baltimore, MD 21205.

E-mail: kchan@jhsph.edu

opyright © 2006 by the Jacobs Institute of Women’s Health.ublished by Elsevier Inc.

emain low to moderate in primary care settingsHarman et al., 2001; Wells et al., 1999; Edlund et al.,004) despite the prevalence of these conditionsYoung et al., 2001; Grant et al., 2004) and efforts tonhance their recognition and management (Rost etl., 2001; Wells et al., 2000). The high prevalence ofepression and anxiety disorders among women

Kessler, 2003) and the extensive functional decrementssociated with these conditions (Kessler et al., 1999)akes the quality of mental health care an important

ealth issue for women (Sherbourne et al., 2001).

It is likely that provider and patient communication

1049-3867/06 $-See front matter.doi:10.1016/j.whi.2006.03.003

sdecpiatetttAmisagdipc

tcaioscWsdcsf

p2ddhc1c(i

qsdiasad(fs(

M

SAdcmec(

T

P

H

M

Q

Aa

Ab

ac

K. S. Chan et al. / Women’s Health Issues 16 (2006) 122–132 123

tyles would influence the detection and treatment ofepression and gender has been implicated in differ-nces in communication styles during clinical en-ounters (Martin et al., 1988). Female primary carehysicians appear to exchange more psychosocial

nformation, engage in more psychosocial counselingnd emotionally focused discussions, and involveheir patients more actively in decision making (Rotert al., 2002; Cooper-Patrick et al., 1999). Female pa-ients are more likely to report physical symptoms ando bring up depression during primary care encoun-ers (Kroenke & Spitzer, 1998; Sleath & Rubin, 2002).lthough these gender differences in communicationay independently influence the clinical encounter, it

s likely that provider and patient communicationtyles also interact to affect what and how informationre exchanged (van den Brink-Muinen et al., 2002). Ifender concordance can be shown to increase rates ofetection and care of mental health problems, match-

ng patients and providers by gender could serve as aractical way to improve the quality of mental healthare for depressed patients.

Gender concordance has been described as impor-ant to women in their relationship with their primaryare physician (Garcia et al., 2003) and has beenssociated with greater trust in a study of physiciansn training (Bonds et al., 2004). However, most studiesn preventive services and counseling in primary careettings have not found positive effects for genderoncordance (Flocke & Gilchrist, 2005; Henderson &

eisman, 2001; Schmittdiel et al., 2000), although 1tudy did find longer visit times for gender-concor-ant patient–provider dyads compared to gender dis-ordant dyads (Franks & Bertakis, 2003). Previoustudies have also found few gender concordance ef-

able 1. Intervention in each QID project

QID Collaborativea

artners-in-Care (PIC) Local clinicians were traineclinicians in their practiceprovide improved medicpsychotherapist trained i

opkins Quality Improvementfor Depression (HQID)

Academic detailing to clinicprogress reports to clinic

ental Health AwarenessProject (MHAP)

Two team-based approachemodel HMOs: 1) a centraacross multiple practices;their own plans.

uality Enhancement byStrategic Teaming (QuEST)

Academic detailing to cliniclocal nurses to educate padjustments over time.

dditional information on the interventions implemented in each sPartners-in-Care (PIC, PI: Kenneth Wells); the Hopkins Quality Imwareness (MHAP, PI: Lisa Rubenstein); and the Quality Enhance

The goal in all 4 projects was to provide high-quality depression carnd Research (Depression Guideline Panel, 1993a, 1993b), while pronsidered to be the most appropriate.

ects on the diagnoses or treatment of psychosocial a

roblems (Schmittdiel et al., 2000; Franks & Bertakis,003; Sleath & Rubin, 2002). However, these studiesid not focus on depressed patients nor control forepression severity. Given that gender differencesave been observed in how patients and providersommunicate (Roter et al., 2002; Kroenke & Spitzer,998; Sleath & Rubin, 2002) and that patient–providerommunication may affect the detection of depressionCarney et al., 1999), it is important to examine thisssue specifically among patients with depression.

To determine whether gender concordance affects theuality of mental health care received in primary careettings, we used a diverse sample of patients with majorepression receiving care during a routine visit to exam-

ne rates of depression detection and mental health carecross different patient–provider gender dyads. We arepecifically interested in whether gender-concordant dy-ds (male–male and female–female) have higher rates ofetection and treatment than gender discordant dyads

male–female and female–male). We examine these ef-ects on depression detection and care, as well as coun-eling for comorbid anxiety and alcohol or drug useKessler, 2003).

ethods

amplesll participants included in this study had majorepression. We examined depression detection andare using baseline data from the Quality Improve-ent for Depression (QID) Collaboration studies (Rost

t al., 2001). Data on anxiety and alcohol or drugounseling were only available from Partners-in-CarePIC), one of the studies participating in QID (Wells et

Intervention Descriptionb

ovide education, academic detailing, and audit/feedback to otherinterventions were developed: 1) local nurses were trained toanagement; 2) patients were able to obtain services from a

itive behavioral therapy for a reduced copay.ase manager to educate and monitor patients and to provide

developed to implement quality improvement plans within staff-anized team of experts developed plans that were implementedlly organized teams at each practice developed and implemented

rained administrative staff to screen patients before visit, trainedand monitor treatment response and provide treatment

s given in the article by Rost et al. (2001).ment for Depression (HQID, PI: Daniel Ford); the Mental Healthy Strategic Teaming (QuEST, PI: Kathryn Rost).fined by recommendations from the Agency for Health Care Policy

g clinicians and patients the freedom to choose the treatment they

d to pr. Two

ation mn cognians; c

ians.s werelly org2) loca

ians, tatients

tudy iprove

ment be, as deovidin

l., 2000). Therefore, we conduct these additional

aa

itpdp4t((Srnaofisspct

I(�micfs

wsepwiQtRrt

rOpptnpWt(t

dpspwd

udeWcgwciTnrQFs

B

Fsittel

K. S. Chan et al. / Women’s Health Issues 16 (2006) 122–132124

nalyses using only data from the PIC sample. For allnalyses, patient data were linked to provider data.QID studies evaluated how different strategies for

mplementing depression practice guidelines affectedhe quality and outcomes of care provided to de-ressed primary care patients (Table 1 provides aescription of the intervention used in each QIDroject) (Rost et al., 2001). This collaboration includesrelated studies with similar designs: the PIC Project;

he Hopkins Quality Improvement for DepressionHQID) Project; the Mental Health AwarenessMHAP) Project; and the Quality Enhancement bytrategic Teaming (QuEST) Project. The QID studiesecruited sites, providers, and patients from 11 orga-izations. These organizations provided both man-ged (staff- and network-model health maintenancerganizations, independent practice associations, pre-erred provider organizations, and public health clin-cs) and nonmanaged (fee-for-service) health care iningle insurer and mixed model clinics. All 4 QIDtudies aimed to recruit a representative patient sam-le and consecutive patient visitors to participatinglinics were approached for screening to determineheir study eligibility.

Eligible patients were positive on the Compositenternational Diagnostic Interview (CIDI) screenerWorld Health Organization [WHO], 1996), indicating2 weeks in the past year and �1 week in the pastonth when they felt sad, empty, depressed, or lost

nterest in things they normally enjoyed, and metriteria for 1-year major depression on a subsequentull structured CIDI interview to be eligible for inclu-ion in the QID core sample.

A baseline interview was subsequently conductedith eligible patients who agreed to participate in the

tudy. Of 2,370 patients who were determined to beligible for the baseline interview, 1,656 (69.9%) com-leted a baseline interview. Of the 1,656, 158 (9.5%)ere subsequently determined to be ineligible for

nclusion in the QID core database, resulting in 1,498ID participants. All QID participants provided writ-

en or verbal informed consent for study participation.esearch protocols were approved by institutional

eview boards at RAND and participating organiza-ions.

We used data from all 4 QID studies to examineates of depression detection and care (Figure 1A).ur QID sample consists of 1,437 patients who com-leted a baseline survey (95.9% of 1,498 QID partici-ants) and 391 providers who provided primary care

o these patients. We excluded 2 providers for whomo gender information was available (0.5% of 391roviders) and their patients (0.6% of 1,437 patients).e retained data from 389 providers and 1,428 pa-

ients for our analyses. Of these 389 providers, 27470.4%) provided personal and professional informa-

ion. Using additional sources, we identified the gen- d

er for the remaining 115 (29.6%) providers. Fiftyercent of our QID sample completed a baselineurvey within 7 days after being screened for majorde-ression. Seventy-five percent were interviewedithin 17 days, 90% within 48 days, 95% within 91ays, and all by 236 days.Data on counseling for anxiety and alcohol or drug

se were available only from PIC. Therefore, we usedata solely from PIC patients with major depression toxamine detection rates for these comorbid conditions.

e included 714 PIC patients and 157 providers whoared for them. Because QID employed a more strin-ent inclusion criterion than PIC (e.g., heavy drinkersere excluded from the QID sample but were in-

luded in the PIC sample), 229 of the 714 PIC patientsncluded in these analyses were PIC-only patients.hat is, they were recruited as PIC participants, butot included as QID participants (Figure 1B). Theelationship between the PIC study sample and theID study sample is presented in Figures 1A and 1B.ifty percent of our PIC sample completed a baselineurvey within 7.5 days after being screened for major

QID Patients (N=1428)

MajorDepression

PICQID Sub-study

Patients(N=485)

MajorDepression

MHAPQID Sub-study

Patients(N=507)

MajorDepression

QuESTQID Sub-study

Patients(N=366)

MajorDepression

HQIDQID Sub-Study

Patients(N=70)

MajorDepression

PIC Patients withMajor Depression

(N=714)

QID Sub-Study Patients (N=485)

Major Depression

Non-QID Patients (N=229)

Major Depression

A

igure 1. (A) The QID sample: distribution of patients across QIDubstudies. (B) The PIC sample: 485 PIC patients were also includedn the QID sample. The 229 patients from PIC were not included inhe QID core database because they either 1) did not expect to usehe plan as their place of care during the next 12 months, 2)xperienced recent bereavement, 3) were heavy drinkers, 4) were onithium, or 5) were currently pregnant.

epression. Seventy-five percent were interviewed

wd

M

Dtdt

bhthcapme

fcttavrm

Ivmfe

C

Dua1arMspts

Aca(

AAs

scbcd

R“mwothdcsLhvdancu

Iampei

AWcdcvgpgrpgbpppf

multap

K. S. Chan et al. / Women’s Health Issues 16 (2006) 122–132 125

ithin 31 days, 90% within 110 days, 95% within 143ays, and all by 236 days.

easures

ependent variables. During the baseline survey, pa-ients reported the actions taken by their providersuring the index visit when they were recruited into

he study.Depression detection. We considered depression to

e detected if patients reported that their providersad either told them that they had depression, treated

hem for depression, or recommended that they getelp for depression. Referral to counseling was indi-ated if providers recommended that patients seenother doctor or therapist for counseling. Medicationrescription was indicated if providers gave patients aedication prescription to help with their personal or

motional problems.Any care. We also constructed a summary indicator

or any care if providers either referred their patient toounseling or prescribed medication. We had addi-ional information from the PIC sample about whetherheir provider talked to them about 1) anxiety and 2)lcohol or drugs during the index visit. All dependentariables are binary and coded so that affirmativeesponse indicates recognition and/or activeanagement.

ndependent variables. We constructed an indicatorariable for each patient–provider gender dyad: 1)ale patients of male providers, 2) male patients of

emale providers, 3) female patients of female provid-rs, and 4) female patients of male providers.

ontrol variables

epression severity. We measured depression severitysing the CESD-23 (score range: 0–100) (Orlando etl., 2000), a variant of the original CES-D (Radloff,977). For the 23-item version, 7 items were removednd 10 new items (See Appendix) added to reflectelevant new criteria from the Diagnostic and Statistical

anual of Mental Disorders (American Psychiatric As-ociation, 1994). We used the overall score to morerecisely account for depressive symptom severity in

he multivariate models. Higher scores indicate moreevere depression in the past month.

nxiety. An anxiety disorder in the past year is indi-ated if patients were classified as having generalizednxiety disorder based on the full CIDI diagnosisWHO, 1996).

t-risk drinking (PIC only). We used scores �8 onUDIT (Alcohol Use Disorders Identification Test;

core range 0–40) (Saunders et al., 1993) to indicate a t

trong likelihood of hazardous or harmful alcoholonsumption in men. Given gender differences inody mass and alcohol metabolism, we used a lowerut-off (scores �7) to enhance sensitivity to at-riskrinking in women (Bradley et al., 1998).

ecent substance use (PIC only). Patients were askedHave you used the following drugs in the past 6onths on your own or nonmedically (i.e., eitherithout a doctor’s prescription or in greater amounts

r more often than prescribed or for a reason otherhan a doctor said you should use them): 1) marijuana,ashish, pot, or grass; 2) barbiturates, sedatives,owners, sleeping pills, seconal, or quaaludes; 3)ocaine, coke, crack, or free base; 4) amphetamines,timulants, uppers, or speed; 5) tranquilizers, Valium,ibrium, or Xanax; 6) heroin; 7) opiates other thaneroin (codeine, Demerol, morphine, methodone, Dar-on, Darvocet, Percodan, Dilaudid, opium); 8) psyche-elics (LSD, mescaline, peyote, psilocybin, DMT, PCP,ngel dust); or 9) inhalants (glue, toluene, gasoline,itrous oxide, amyl nitrate poppers).” Patients wereonsidered current substance users if they reportedsing any of these substances.

ntervention status. We created indicators to control forny intervention effect on patients and providers whoay have been exposed to the intervention when

atients took the baseline survey. The intervention forach of the QID projects is described in greater detailn Table 1.

nalysise used t-tests and �2 tests to compare provider

haracteristics by provider gender. We linked patientata to provider data to examine rates of detection andare. We examined rates by patient gender and pro-ider gender as well as across the 4 patient–providerender dyads. We initially tested for differences byatient gender and provider gender separately and forender concordance effects using unadjusted logisticegression models. For our gender concordance com-arisons, we constructed dummy variables for theender dyad groups and performed specific testsetween concordant and discordant patient–providerairs (i.e., if female patients do better with femaleroviders than with male providers, and if maleatients do better with male providers than with

emale providers).We then tested the robustness of our findings usingultivariate logistic regression models that control for

nderlying need (CES-D 23 score, anxiety status, prob-em drinking, substance use), intervention status, pa-ient age, patient ethnicity, patient education, providerge, number of years provider was in practice, androvider practice specialty (general or family practi-

ioners; residents or nurse–practitioners; internists).

Wmmemgb

stwi(fdadp

sdaipifSwecjaf

R

PFwtfp(A(ppnp�plpfc

ps

POwasgdwsus1ssafeo

RArHcm2arapwPbg

sapdFacsrvscdrmst

K. S. Chan et al. / Women’s Health Issues 16 (2006) 122–132126

e examined main gender effects on our outcomes byodeling patient gender and provider gender in theseultivariate logistic regression models. Finally, we

xamined the effects of gender concordance in theseultivariate models using dummy variables for the

ender dyad groups and performing specific testsetween gender-concordant and -discordant pairs.Finally, because the Veterans Administration (VA)

ample (n � 105) had no female patients and becausehe VA differs significantly from other systems of care,

e conducted a sensitivity analysis. We added 2ndicator variables, 1 for each of the 2 VA groupsmale patients of male providers and male patients ofemale providers) to our models to determine whetherifferences across the 4 non-VA gender groups remainfter VA patients are separately modeled. This wasone only for analyses using the QID sample as VAatients were not part of the PIC project.The proportion of missing patient data in our study

ample is low, ranging between 0% and 10% forifferent variables. Nevertheless, to limit sample lossnd bias from case-wise deletion, we used multiplemputation for data missing at the item level. Theredicted outcomes were averaged across 5 randomly

mputed data sets and standard errors were adjustedor the uncertainty in the imputations ( Rubin, 1987;chafer, 1997). Data for all patient-level analyses areeighted for probability of enrollment to the full

ligible sample. Descriptive comparisons of providerharacteristics were not weighted. Our models ad-usted for clustering of patients within providers using

modified robust variance estimator (Bell & McCaf-rey, 2002).

esults

rovider Characteristicsorty-one percent (n � 158) of the 389 QID providersere female. Of the 274 providers who reported on

heir background, 39% were internists, 43% general oramily practitioners, and 18% residents or nurse–ractitioners. They were, on average, 43.4 years old

SD � 8.3) and had practiced for 11.7 years (SD � 7.7).higher proportion of male providers were internists

44% versus 32%, p � .04) and general or familyractitioners (51% versus 32%, p � .01); a higherroportion of female providers were residents orurse–practitioners (36% versus 5%, p � .01). The maleroviders were older, with an average age of 45.6 (SD8.5) years relative to 40.2 (SD � 6.9) years for female

roviders (p � .01). Male providers had also practicedonger, with an average of 14.4 (SD � 8.3) years ofractice relative to 7.9 (SD � 4.6) years of practice in

emale providers (p � .01). The distribution of spe-

ialty, age, and years of practice for the 157 PIC t

roviders (36% female), overall and by gender, wasimilar to QID.

atient Characteristicsf the 1,428 QID patients, 71% were female, 67% werehite, and 57% attended some college, with an aver-

ge age of 44.6 years (Table 2). Racial/ethnic compo-ition and depression severity were comparable acrossender-concordant and gender-discordant dyads, butifferences in patient age and educational attainmentere observed. The PIC sample was demographically

imilar to the QID sample (Table 2). For variablesnique to PIC, we found that 54% of our PIC samplecreened positive for generalized anxiety disorder,4% were at-risk drinkers, and 30% reported recentubstance use. Race/ethnic composition, depressioneverity, and recent substance use were comparablecross gender-concordant/-discordant dyads, but dif-erences in patient age, educational attainment, pres-nce of anxiety disorder, and at-risk drinking werebserved.

ates of Depression Detection and Caremong QID patients with major depression, only 33%

eported that their provider detected their depression.owever, 40% reported receiving either a referral for

ounseling and/or a medication prescription, whichay reflect poor recall by some patients. Specifically,

0% of patients reported being referred to counselingnd 30% being given prescriptions, with some patientseceiving both. Reported rates of detection and care,cross the various indicators, were comparable acrossrovider gender and patient gender (Table 3). Ratesere also similar across gender dyads (Table 4).atients with a provider of concordant gender did noetter than patients with a provider of discordantender.Controlling for depression severity, intervention

tatus, and patient and provider characteristics did notffect results (Table 5). Neither patient gender norrovider gender was significantly related to any of 4epression care outcomes in our multivariate models.urthermore, after adjustment for these control vari-bles, discordant gender groups did no worse thanoncordant gender groups. Male patients who wereeen by a female provider were not less likely to reporteceiving depression care as male patients with pro-ider of the same gender. Similar results were ob-erved for female patients. Depression detection andare remained comparable across the non-VA genderyads even when the VA gender dyads were sepa-ately modeled. Intervention status, from the adjustedodels, was not significantly associated with depres-

ion detection and care (all p � .10), probably becausehe diffusion of the interventions was limited at the

ime of the baseline survey.

DROat

no3oH

T

Q

P

NA

T

T

R

TT

Na

b

c

K. S. Chan et al. / Women’s Health Issues 16 (2006) 122–132 127

etection of Anxiety Disorderates of anxiety detection in our PIC sample were low.nly 25% of patients in this sample who had gener-

lized anxiety disorder in the past year reported thatheir provider talked to them about anxiety. No sig-

able 2. Patient sample characteristics

ID Collaboration

Total Sample

MAll Patientsn � 1,428

WeightedMean or

%

Age, mean, SD 1,428 44.6 (13.7)Race

White 990 67%Black 126 9%Hispanic 198 15%Other 114 8%

Education� High school 167 13%High school 408 30%Some college 523 35%College 330 22%

CES-D score (0–100), mean, SD 1,428 50.1 (20.9)

artners-in-Care n � 714

Age, mean, SD 714 43.2 (14.2)Race

White 450 60%Black 40 6%Hispanic 178 28%Other 46 6%

Education� High school 96 17%High school 191 30%Some college 259 33%College or more 168 20%

CES-D score (0–100), mean, SD 714 49.2 (19.1)Generalized anxiety disorder,

past year, % yes714 54%

Problem drinking (based onAUDIT), % yes

714 14%

Any substance use, past 6 mo,% yes

714 30%

ote: QID: 231 male providers, 158 female providers; PIC: 100 malebbreviations: QID, Quality Improvement for Depression.

able 3. Rates of detection or management of depression and coun

old patient they had depression, treated them orrecommended they get help for depression

eferred to counseling or received medication prescriptionReferral to physician or therapist for counselingReceived medication for personal or emotional problems

alked to patient about anxietya

alked to patient about alcohol or drugsa

ote: Unadjusted percentages, statistical tests are based on a modeData from Partners-in-Care major depression sample.Ho: male provider � female provider, p � 0.05.

Ho: male patient � female patient, p � 0.05.

ificant differences in counseling for anxiety werebserved by provider gender or patient gender (Table). Similar rates in counseling for anxiety were alsobserved across the four gender dyads (Table 4).owever, after adjustment for depression symptoms,

atientrovider237

Male PatientFemale Provider

n � 130

Female PatientMale Provider

n � 580

Female PatientFemale Provider

n � 481

(14.7) 50.0 (13.8) 44.1 (13.5) 41.6 (12.5)

% 68% 67% 66%% 9% 10% 9%% 11% 15% 17%% 12% 7% 8%

% 8% 17% 11%% 28% 34% 28%% 43% 32% 37%% 21% 18% 24%(20.5) 51.6 (22.4) 51.7 (20.7) 48.5 (20.7)

132 n � 39 n � 301 n � 242

4 (14.7) 43.3 (14.9) 44.1 (14.7) 40.0 (12.4)

% 56% 58% 59%% 9% 4% 8%% 27% 32% 29%% 8% 6% 5%

% 11% 23% 10%% 34% 31% 29%% 37% 30% 37%% 17% 16% 24%3 (18.6) 52.3 (16.3) 49.8 (18.9) 48.3 (20.1)% 50% 59% 49%

% 28% 12% 11%

% 31% 32% 25%

ders, 57 female providers.

for comorbid conditions by provider gender and patient gender

Provider Gender Patient Gender

ale Female Male Female

4% 32% 31% 34%

0% 39% 40% 40%8% 22% 22% 19%2% 27% 27% 31%6% 19% 19% 17%8%b 2%b 12%c 3%c

g main effects for provider gender and patient gender.

Male Pale Pn �

47.7

707

157

12253329

48.9

n �

46.

695

206

17282926

48.51

21

34

provi

seling

M

3

4131

l testin

apl(dtvdd

DTwtsv

sswapwgsdtecspwb

Tp

T

R

TT

Na

b t–male

T

MPPG

M

F

a

pi**

K. S. Chan et al. / Women’s Health Issues 16 (2006) 122–132128

nxiety status, intervention status, and patient androvider characteristics, female providers were more

ikely to counsel their depressed patients on anxietyTable 5). Furthermore, female patients with a gender-iscordant provider was less likely to be counseled

han female patients with a gender-concordant pro-ider, whereas male patients with a gender-concor-ant provider did not differ from those with a gender-iscordant provider (Table 5).

etection of Alcohol or Drug Use Problemshe rate of detection for alcohol or drug use problemsas particularly poor in our PIC sample. Only 9% of

hose who were at-risk drinkers or indicated recentubstance use reported being counseled by their pro-ider on alcohol or drug use. Unlike care for depres-

able 4. Percentage of patients reporting detection or managemenatient gender dyads

old patient they had depression, treated them, orrecommended they get help for depression

eferred to counseling or received medication prescriptionReferral to physician or therapist for counselingReceived medication for personal or emotional problems

alked to patient about anxietya

alked to patient about alcohol or drugsa

ote: Unadjusted percentages; statistical tests are based on an unadData from Partners-in-Care major depression sample.Ho: Female patient–female provider (concordant) � female patien

able 5. Adjusted odds ratio of patient gender, provider gender, a

QID Sample

Talked aboutor ManagedDepression

Adjusted OR(95% CI)

PatientReferred to

Counseling orReceived

MedicationAdjusted OR

(95% CI)

odel: main gender effectsa

atient female gender 1.00 (0.65–1.54) 0.85 (0.61–1.17)rovider female gender 1.03 (0.68–1.55) 1.14 (0.79–1.64)ender concordancecomparisonsa

ale patients: discordantvs. concordant providergender (reference)

0.90 (0.41–1.97) 1.07 (0.55–2.09)

emale patients: discordantvs. concordant providergender (reference)

0.94 (0.61–1.46) 0.87 (0.59–1.27)

Both models control for patient characteristics (age, ethnicity, eduractice), and intervention status; anxiety status is added to PIC m

ndicator for recent substance use are added to the PIC model on cp � 0.05.

*p � 0.01.

ion or anxiety, we did find gender effects for coun-eling on alcohol or drugs (Table 3). Female providersere less likely than male providers to counsel on

lcohol or drug use (2% versus 8%, p � .03) and femaleatients were less likely to be counseled comparedith male patients (3% versus 12%; p � .01). However,

ender concordance did not necessarily increase coun-eling (Table 4). In fact, female patients with gender-iscordant providers reported higher counseling rates

han female patients with gender-concordant provid-rs (5% versus 1%; p � .04). Male patients of gender-oncordant providers did report higher levels of coun-eling than male patients of gender-discordantroviders (14% versus 5%). However, this differenceas not statistically significant (p � .31), possibly

ecause of the small sample size in the male patient–

ression and counseling for comorbid conditions by provider and

Male Patient Female Patient

alevider

FemaleProvider

MaleProvider

FemaleProvider

1% 32% 36% 32%

0% 40% 40% 39%9% 26% 18% 21%0% 23% 32% 29%9% 19% 14% 19%4% 5% 5%b 1%b

model.

provider (discordant), p � 0.05.

der concordance from multivariate models

,428) PIC Sample (n � 714)

rral tonselingsted OR% CI)

ReceivedMedication for

EmotionalProblems

Adjusted OR(95% CI)

Counseling forAnxiety

Adjusted OR(95% CI)

Counseling forAlcohol or

Substance UseAdjusted OR

(95% CI)

.64–1.45) 0.91 (0.65–1.28) 0.66 (0.34–1.28) 0.28 (0.10–0.77)*

.83–1.98) 1.07 (0.73–1.57) 1.81 (1.20–2.73)** 0.16 (0.03–0.80)*

.55–2.63) 0.90 (0.43–1.90) 1.06 (0.38–2.93) 0.16 (0.01–2.24)

.47–1.24) 0.89 (0.60–1.32) 0.49 (0.27–0.90)* 6.22 (1.15–33.6)*

CES-D 23 score), provider characteristics (age, specialty, years inon counseling for anxiety, while problem drinking status and aning for alcohol or substance use.

t of dep

MPro

3

41311

justed

nd gen

(n � 1

RefeCou

Adju(95

0.96 (01.28 (0

1.20 (0

0.77 (0

cation,odel

ounsel

fwwcr

racmlasccsvcpaeppw

D

Rcmdcdl2w(cwftd

asdswmphmscia

mcvpalppdf

gumdMiplarcgpt(2pmsmdn(cpmheb(ncgr

apttgsdSSc

K. S. Chan et al. / Women’s Health Issues 16 (2006) 122–132 129

emale provider group (n � 39). The largest differenceas observed between the 2 gender-concordant groups,ith male patients of male providers reporting the most

ounseling and female patients of female providerseporting the fewest (14% versus 1%; p � .01).

Adjusting for depression severity, at-risk drinking,ecent substance use, intervention status, and patientnd provider characteristics did not substantiallyhange the pattern of findings from the unadjustedodels (Table 5). Female patients were significantly

ess likely than male patients to receive counseling onlcohol and substance use. Female providers wereignificantly less likely than male providers to offerounseling on alcohol and substance use. Genderoncordance did not increase the likelihood of coun-eling. Female patients with gender-discordant pro-iders were actually more likely to get counselingompared to female patients with gender-concordantroviders. The likelihood of receiving counseling forlcohol or substance use was not significantly differ-nt between male patients with gender-concordantroviders and male patients with gender-discordantroviders, despite lower odds among male patientsith gender-discordant providers.

iscussion and Conclusions

ates of detection and care for depression in primaryare settings were low even among patients withajor depression. Only 30%–40% of our sample,

rawn from all 4 collaborative studies in QID, indi-ated that their provider recognized or dealt with theirepression at their screening visit. These results rep-

icate those from an earlier PIC-only study in which9% of those with single depression and 43% of thoseith double depression receive any depression care

Wells et al., 1999). Detection rates for comorbidonditions were even lower. Only 25% of our sampleith comorbid anxiety and 9% of our sample at-risk

or drinking or substance abuse problems reportedhat their provider talked to them about these issuesuring their visit.We found little evidence that gender concordance

ffects the detection and care of mental health andubstance use problems for depressed patients. Gen-er concordance was expected to enhance self-disclo-ure and trust (Bonds et al., 2004), which in turnould facilitate the diagnosis and management ofental health problems. However, the percentage of

atients in our study who reported that their doctorsad talked to them about their depression or anxiety,ade referrals, or provided medications for depres-

ion were strikingly similar for patients from gender-oncordant and gender-discordant dyads. Differencesn counseling on alcohol and drugs were observed

cross gender dyads, but these differences were due to m

ain provider and patient gender effects, not genderoncordance. Adjustment for underlying need, inter-ention status, and characteristics of patients androviders did not change our findings, except fornxiety counseling in female patients who were moreikely to receive counseling if they had a femalerovider. Overall, our findings suggest that matchingatients and primary care providers by gender wouldo little to enhance the quality of mental health care

or depressed patients.We found few main effects of provider or patient

ender on depression or anxiety counseling. Duringsual primary care encounters, similar percentages ofale and female patients reported that their providers

etected and cared for their depression and anxiety.ale and female providers also do not appear to differ

n their rates of detection and treatment of theiratients’ depression, even after adjustment for under-

ying need, patient characteristics, and provider char-cteristics. However, adjustment for these variablesevealed that female providers were more likely toounsel on anxiety. The lack of a general providerender effect was surprising, given that higher rates ofreventive and counseling services have been consis-

ently reported among patients of female providersFlocke & Gilchrist, 2005; Henderson & Weisman,001; Lurie et al., 1997). However, a positive femalerovider effect has been less consistently reported forental health detection and counseling. Although 1

tudy found that female primary care providers wereore likely to counsel their patients on anxiety and

epression (Sleath & Rubin, 2002), other studies foundo provider gender effect on psychiatric diagnosesFranks & Bertakis, 2003) or counseling on psychoso-ial issues (Schmittdiel et al., 2000). The lack of aatient gender effect was also surprising, given thatale patients are less likely to have their mental

ealth problems detected (Borowsky et al., 2000; Pottst al., 1991) and female patients were more likely toring up depression during the clinical encounterSleath & Rubin, 2002). However, it is important toote that these studies included patients withouturrent major depression and it is possible that patientender effects diminish as symptoms become moreeadily recognizable with greater depression severity.

We did find sizable gender effects for counseling onlcohol or drug use, with male patients and maleroviders more likely to engage in counseling than

heir female counterparts, even after accounting forhe control variables in our multivariate model. Theseender effects may stem from the fact that alcohol andubstance use is higher and has been linked withepression among men (National Center for Healthtatistics, 2003; Substance Abuse and Mental Healthervices Administration, 2003; Kessler, 2003), whichould prompt providers to screen for these disorders

ore actively in male patients. Meanwhile, the social

sefwpomeefwmiprapva

sdpotelttpntdasippnrdsptpmbtatuHtcpsm

tcttwtspwceho

ctprsafadfftpscs

AS(RCHaCMCCJKa

patMOAtutTH

K. S. Chan et al. / Women’s Health Issues 16 (2006) 122–132130

tigma of substance use for women may make provid-rs more hesitant in probing the same issues with theiremale patients (Monroe et al., 1997). It is less clear

hy male providers were more likely to talk to theiratients about alcohol or drug use. However, a previ-us study has also found that male physicians devotedore time to discussions of substance abuse (Bertakis

t al., 2003). It is possible that this provider genderffect is also attributable to the fact that men do notind alcohol and substance use as stigmatizing as

omen do and therefore male providers have fewerisgivings about raising these topics during the clin-

cal encounter. Regardless, our findings suggest thatrimary care providers should be encouraged to moreoutinely and systematically screen for developinglcohol and substance use problems in their depressedatients, particularly because brief counseling inter-entions have been shown to be effective in both mennd women (Manwell et al., 2000; Wilk et al., 1997).Our study has a number of limitations. First, our

ample was not distributed evenly across the 4 genderyad groups and small sample sizes in some cells,articularly for the PIC sample, may have influencedur findings. In particular, the number of male pa-ients with female providers was small and may havexaggerated the magnitude of the effect as well asimited our ability to detect differences in comparisonshat include this group. Our findings for comparisonshat include cells with small samples should be inter-reted cautiously. Second, given the observationalature of our study, important factors not examined in

his study may affect our findings. For example, theuration of physician–patient relationship may inter-ct with gender, with gender effects exerting theirtrongest influence early in the relationship but dimin-shing over time as patients develop trust with theirrovider. Third, although we were able to assessatient-reported rates of detection and care, we didot confirm patient reports with providers or medicalecords and we lacked sufficient data to examine inetail the quality of the care provided during thepecific visit. Fourth, our sample is composed ofatients who agreed to participate in a randomized

rial and may be more homogeneous than depressedrimary care patients in general. Fifth, although theental health problems examined in this study tend to

e recurrent or chronic, it is possible that some pa-ients screened for depression in the past month or annxiety disorder in the past year may not be symp-omatic at the index visit. Therefore, we may havenderestimated the rate of detection by providers.owever, it is important that providers talk to even

heir asymptomatic patients about their mental healthoncerns, particularly if they are aware that theiratients have experienced a recent episode of depres-ion or anxiety or that the patient has a history of

ental health problems. Finally, if patients reported L

hat their providers talked to them about a topic, weonsidered that counseling. However, we do not knowhe extent to which counseling was actually provided;hus, our counseling variables might include instances

hen the topic was only mentioned during historyaking. However, our study does offer a valuablenap-shot of the mental health care that depressedatients receive during routine primary care visits inhich depression must compete with other health

oncerns for care. In addition, the geographic andthnic diversity of our samples and the variety ofealth systems included enhance the generalizabilityf our findings.In summary, we found little evidence that gender

oncordance increases the detection and care of men-al health and substance use problems for depressedatients in the primary care setting. However, the lowates of detection and treatment for mental health andubstance use problems are of concern and warrantdditional efforts to improve care of both male andemale patients with major depression. Our findingslso indicate that gender can affect whether problemrinking and substance abuse are identified and cared

or. More research is needed to understand whyemale providers and female patients are less likely toalk about substance abuse. However, encouragingrimary care providers to routinely and systematicallycreen for these problems may enhance this aspect ofare for both men and women suffering from depres-ion.

cknowledgmentsupported by the National Institutes of Mental HealthR01MH64658, P50MH54623), the Agency for Healthcareesearch & Quality (R01-HS08349), and the John D. andatherine T. MacArthur Foundation, Grant No. 96-42901A-E. We thank Bernadette Benjamin for expert programming

nd we are grateful to Christy Klein, Maureen Carney, andhantal Avila for help in the PIC study; Carole Oken andary Abdun-Nur for help in the MHAP study; Jeff Smith,arl Elliot, and Paul Nutting for help in the QuEST study;hristine Nelson, Ray Turner, Tracey Hare, Hong Vu, and

ose Arbelaez for help in HQID and the combined database;athryn Magruder for help with the overall collaboration;nd Bob Bell for statistical consultation.We thank the organizations participating in Quality Im-

rovement for Depression (QID) studies, who providedccess to their expertise and patients, implemented interven-ions, and provided in-kind resources: Kaiser Permanente

edical Care Programs in Northern California Region,akland, CA; VA Greater Los Angeles, Los Angeles, CA;mbulatory Sentinel Practice Network, Denver, CO, a prac-

ice-based research network of family physicians that vol-ntarily participates in research; NYL-Care Health Plans of

he Mid-Atlantic, Greenbelt, MD; Allina Medical Group,win Cities, MN; Columbia Medical Plan, Columbia, MD;umana Health Care Plans, San Antonio, TX; MedPartners,

os Angeles, CA; PacifiCare of Texas, San Antonio, TX;

VioTw(c

At

INI

III

II

II

RA

B

B

B

B

B

C

C

D

D

E

F

F

G

G

H

H

K

K

K

L

M

M

M

N

O

P

R

R

R

K. S. Chan et al. / Women’s Health Issues 16 (2006) 122–132 131

alley-Wide Health Services, Alamosa, CO; Patuxent Med-cal Group (MD); and to their associated behavioral healthrganizations: Alamo Mental Health Group (San Antonio,X), San Luis Valley Mental Health/Colorado Health Net-orks (CO), and Magellan/GreenSpring Behavioral Health

MD). We are also grateful to the clinicians and patients whoontributed their time and efforts for these studies.

ppendix. Item content added to and deleted fromhe CESD-20 in the current CESD-23 version

Items Added Items Deleted

considered suicideothing was fun or enjoyablehad less interest in my usualactivities

was eating way too muchslept too muchmoved much slower thannormal

was restless or jitteryblamed myself for the waythings are

could not concentratethought a lot about death

I was bothered by things thatusually don’t bother me

I felt people disliked meI thought my life had been a

failureI felt fearfulI talked less than usualI felt lonelyPeople were unfriendly

eferencesmerican Psychiatric Association. (1994). Diagnostic and statisticalmanual of manual disorders (4th ed.). Washington, DC: Author.

ell, R. M., & McCaffrey, D. F. (2002). Bias reduction in standarderrors for linear regression with multi-stage samples. SurveyMethodology, 28(2), 169–181.

ertakis K. D., Franks P., & Azari R. (2003). Effects of physiciangender on patient satisfaction. Journal of the American MedicalWomens Association, 58, 69–75.

onds, D. E., Foley, K. L., Dugan, E., Hall, M. A., & Extrom, P.(2004). An exploration of patients’ trust in physicians in training.Journal of Health Care for the Poor and Underserved, 15, 294–306.

orowsky, S. J., Rubenstein, L. V., Meredith, L. S., Camp, P.,Jackson-Triche, M., & Wells, K. B. (2000). Who is at risk ofnondetection of mental health problems in primary care? Journalof General Internal Medicine, 5, 381–388.

radley, K. A., Boyd-Wickizer, J., Powell, S. H., & Burman, M. L.(1998). Alcohol screening questionnaire in women: a criticalreview. Journal of the American Medical Association, 280, 166–171.

arney, P. A., Eliassen, M. S., Wolford, G. L., Owen, M., Badger,L. W., & Dietrich, A. J. (1999). How physician communicationinfluences recognition of depression in primary care. Journal ofFamily Practice, 48, 958–964.

ooper-Patrick, L., Gallo, J. J., Gonzales, J. J., Vu, H. T., Powe, N. R.,Nelson, C., et al. (1999). Race, gender, and partnership in thepatient-physician relationship. Journal of the American MedicalAssociation, 282, 583–589.

epression Guideline Panel (1993a). Depression in primary care. Vol.1: Detection and diagnosis. Clinical practice guideline. no. 5 (AH-CPR Publication No. 93-0550). Rockville, MD: U.S. Department ofHealth and Human Services; Public Health Service; Agency forHealth Care Policy and Research.

epression Guideline Panel (1993b). Depression in primary care. Vol.2: Treatment of major depression. Clinical Practice Guideline. No. 5(AHCPR Publication No. 93-0551). Rockville, MD: U.S. Depart-ment of Health and Human Services; Public Health Service;

Agency for Health Care Policy and Research.

dlund, M. J., Unutzer, J., & Wells, K. B. (2004). Clinician screeningand treatment of alcohol, drug and mental problems in primarycare: Results from Healthcare for Communities. Medical Care, 42,1158–1166.

locke, S. A., & Gilchrist, V. (2005). Physician and patient genderconcordance and the delivery of comprehensive clinical preven-tive services. Medical Care, 43, 486–492.

ranks, P., & Bertakis, K. D. (2003). Physician gender, patientgender, and primary care. Journal of Women’s Health, 12, 73–80.

arcia, J. A., Paterniti, D. A., Romano, P. S., & Kravitz, R. L. (2003).Patient preferences for physician characteristics in university-based primary care clinics. Ethnicity & Disease, 13, 259–267.

rant, B. F., Stinson, F. S., Dawson, D. A., Chou, S. P., Dufour, M. C.,Compton, W., et al. (2004). Prevalence and co-occurrence ofsubstance use disorders and independent mood and anxietydisorders. Archives of General Psychiatry, 61, 807–816.

arman, J. S., Schulberg, H. C., Mulsant, B. H., & Reynolds, C. F., III.(2001). The effect of patient and visit characteristics on diagnosisof depression in primary care. Journal of Family Practice, 50, 1068.

enderson, J. T., & Weisman, C. S. (2001). Physician gender effectson preventive screening and counseling: An analysis of male andfemale patients’ health care experiences. Medical Care, 39, 1281–1292.

essler, R. C. (2003). Epidemiology of women and depression.Journal of Affective Disorders, 74, 15–13.

essler, R. C., DuPont R. L., Berglund, P., & Wittchen, H. U. (1999).Impairment in pure and comorbid generalized anxiety disorderand major depression at 12 months in two national surveys.American Journal of Psychiatry, 156, 1915–1923.

roenke, K., & Spitzer, R. L. (1998). Gender differences in thereporting of physical and somatoform symptoms. PsychosomaticMedicine, 60, 150–155.

urie N., Margolis K. L., McGovern P. G., Mink, P. J., & Slater J. S.(1997). Why do patients of female physicians have higher rates ofbreast and cervical cancer screening. Journal of General InternalMedicine, 12, 34–43.

anwell, L. B., Fleming, M. F., Mundt, M. P., Stauffacher, E. A., &Barry, K. L. (2000). Treatment of problem alcohol use in womenof childbearing age: Results of a brief intervention trial. Alcohol-ism, Clinical and Experimental Research, 24, 1517–1524.

artin, S. C., Arnold, R. M., & Parker, R. M. (1988). Gender andmedical socialization. Journal of Health and Social Behavior, 29,333–343.

onroe, A. D., Levy, S., & McQuade, W. (1997). Effects of gender ondocumentation of alcohol use in patients with psychiatric symp-toms. Substance Abuse, 18, 79–87.

ational Center for Health Statistics (2003). Health, united states,2003, with chartbook on trends in the health of Americans. Hyattsville,MD: Author.

rlando, M., Sherbourne, C. D., & Thissen, D. (2000). Summed-scorelinking using item response theory: Application to depressionmeasurement. Psychological Assessment, 12, 354–359.

otts, M. K., Burnam, M. A., & Wells, K. B. (1991). Genderdifferences in depression detection: A comparison of cliniciandiagnosis and standardized assessment. Psychological Assessment,3, 609–615.

adloff, L. S. (1977). The CES-D scale: A self-report depression scalefor research in the general population. App Psych Measurement, 1,385–401.

ost, K. M., Duan, N., Rubenstein, L. V., & et al., for the QualityImprovement for Depression (QID) Consortium. (2001). TheQuality Improvement for Depression Collaboration: Generalanalytic strategies for a coordinated study of quality improve-ment in depression care. General Hospital Psychiatry, 23, 239–253.

oter, D. L., Hall, J. A., & Aoki, Y. (2002). Physician gender effectsin medical communication: A meta-analytic review. Journal of the

American Medical Association, 288, 756–764.

R

S

S

S

S

S

S

S

v

W

W

W

W

Y

K. S. Chan et al. / Women’s Health Issues 16 (2006) 122–132132

ubin, D. B. (1987). Multiple imputation for nonresponse in surveys.New York: John Wiley & Sons.

aunders, J. B., Aasland, O. G., Babor, T. F., de la Puente, J. R., &Grant, M. (1993). Development of the Alcohol Use DisordersScreening Test (AUDIT): WHO collaborative project on earlydetection of persons with harmful alcohol consumption. II.Addiction, 88, 791–804.

chafer, J. L. (1997). Analysis of incomplete multivariate data. London:Chapman & Hall.

chmittdiel, J., Grumbach, K., Selby, J. V., & Quesenberry, C. P.(2000). Effect of physician and patient gender concordance onpatient satisfaction and preventive care practices. Journal ofGeneral Internal Medicine, 15, 761–769.

choenbaum, M., Unützer, J., McCaffrey, D., Duan, N., Sherbourne,C., & Wells, K. (2002). The effects of primary care depressiontreatment on patient’s clinical status and employment. HealthServices Research, 37, 1145–1158.

herbourne, C. D., Dwight-Johnson, M., & Klap, R. (2001). Psycho-logical distress, unmet need, and barrier to mental health care forwomen. Women’s Health Issues, 11, 231–243.

leath, B., & Rubin, R. H. (2002). Gender, ethnicity, and physician-patient communication about depression and anxiety in primarycare. Patient Education and Counseling, 48, 243–252.

ubstance Abuse and Mental Health Services Administration (2003).

Results from the 2002 National Survey on Drug Use and Health:

National findings [NHSDA series H-22, DHHS publication no.SMA 03-3836]. Rockville, MD: Office of Applied Studies.

an den Brink-Muinen, A., van Dulmen, S., Messerli-Rohrbach, V.,& Bensing, J. (2002). Do gender dyads have different communi-cation patterns? A comparative study in Western-Europe generalpractices. Patient Education and Counseling, 48, 253–264.ells, K. B., Schoenbaum, M., Unützer, J., Lagomasino, I. T., &Rubenstein, L. V. (1999). Quality of care for primary care patientswith depression in managed care. Archives of Family Medicine, 8,529–536.ells, K. B., Sherbourne, C., Schoenbaum, M., Duan, N., Meredith,L., Unutzer, J., et al. (2000). Impact of disseminating qualityimprovement programs for depression in managed depressioncare: A randomized controlled trial. Journal of the AmericanMedical Association, 283, 212–220.ilk, A. I., Jensen, N. M., & Havighurst, T. C. (1997). Meta-analysisof randomized control trials addressing brief interventions inheavy alcohol drinkers. Journal of General Internal Medicine, 12,274–283.orld Health Organization (WHO). (1996). Composite internationaldiagnostic interview for primary care. Version 2.0. Geneva: Author.

oung, A. S., Klap, R., Sherbourne, C. D., & Wells, K. B. (2001). Thequality of care for depressive and anxiety disorders in the United

States. Archives of General Psychiatry, 58, 55–61.