General Hospital Psychiatry 25 (2003) 158  168
Improving primary care treatment of depression among patients with
diabetes mellitus: the design of the Pathways Study
Wayne Katon, M.D.a,*, Michael Von Korff, ScD.b, Elizabeth Lin, M.D., M.P.H.b,
Greg Simon, M.D., M.P.H.b, Evette Ludman, Ph.D.b, Terry Bush, Ph.D.b, Ed Walker, M.D.a,
Paul Ciechanowski, M.D., M.P.H.a, Carolyn Rutter, Ph.D.b
a
Department of Psychiatry and Behavioral Sciences, University of Washington School of Medicine, Seattle, WA 98195 6580, USA
b
Center for Health Studies, Group Health Cooperative, Psychiatry Service, University Hospital, Seattle, WA 98195 6580, USA
Abstract
This paper describes the methodology of a population based study of primary care patients with diabetes mellitus enrolled in a health
maintenance organization. The first goal was to determine the prevalence and impact of depression in patients with diabetes. The second
goal was to randomize approximately 300 patients with diabetes and major depression and/or dysthymia in a trial to test the effectiveness
of a collaborative care intervention in improving quality of care and health outcomes among patients with diabetes and depression. © 2003
Elsevier Inc. All rights reserved.
1. Introduction Both major depression and depressive symptoms have been
shown in most studies to be associated with glucose dysregu-
Diabetes is a common and costly condition that affects 16
lation [6 8]. This may either be due to the adverse impact of
million Americans [1]. Patients with diabetes are at increased
depression on diabetes self-care (i.e., diet, exercise, checking
risk of kidney disease, peripheral vascular disease, heart dis-
blood glucose, and taking medications), [6 8] direct adverse
ease, lower extremity ulcers and amputations, retinal disease,
physiologic effects on glucose metabolism, [9] or a combina-
neuropathy, infections, digestive disease and periodontal dis-
tion of these two mechanisms. Several studies have shown that
ease [2]. The prevalence is as high as 20% in patients who are
depression is associated with poor adherence to self-care reg-
65 years and over with significantly higher rates in minority
imens such as checking blood glucose, following a special diet
group members (African Americans, Hispanics and Native
and medication compliance, as well as less sensitivity to insu-
Americans) [3]. The total direct medical costs and indirect
lin effects on lowering blood glucose [6,10].
costs in the United States due to diabetes have been estimated
Two small randomized trials have shown that nortripty-
at $102 billion per year [4]. Patients with diabetes appear to
line [11] and fluoxetine [12] were more efficacious in con-
have increased risk of major depression. Depression may ad-
trolling depressive symptoms than placebo in patients with
versely affect self-care regimens as well as increase risk of
major depression and diabetes. One cognitive behavioral
complications such as diabetic retinopathy [5 7].
trial also demonstrated enhanced efficacy compared to an
Anderson and colleagues recent meta-analysis of the
educational diabetes group [13]. One of these 3 trials found
prevalence of major depression or depressive symptoms in
that improved depression outcomes was associated with
patients with diabetes found a two-fold higher prevalence
improved HbA1C levels [13]. These three trials enrolled a
rate of depression in diabetics compared to controls in 20
combined total of less than 180 patients and lacked power to
controlled studies [5]. Current major depression was ob-
study key outcomes such as the effect of enhanced treatment
served in 10 to 15% of diabetics with similar prevalence
of depression on self-care regimens (diet, exercise, refilling
rates in Type 1 and Type 2 cases [5].
medication), disability, quality of life, and medical costs.
Whether enhanced treatment of depression improves glyce-
mic control is also an unanswered question.
* Corresponding author. Tel.: [phone]; fax: [phone]-
The study that we describe has two arms: 1) a popula-
5414.
E-mail address: [e-mail] (W.J. Katon). tion-based epidemiologic investigation of the prevalence
0163-8343/03/$  see front matter © 2003 Elsevier Inc. All rights reserved.
doi:10.1016/S0163-8343(02)00013-6
W.J. Katon et al. / General Hospital Psychiatry 25 (2003) 158  168 159
and impact of depression in patients with diabetes enrolled 2.2. Sample recruitment
in a health maintenance organization; and 2) a randomized
A major methodologic issue was how to best screen and
controlled trial to test the effectiveness of collaborative care
enroll a representative sample of diabetic patients with ma-
interventions in improving the quality of care and outcomes
jor depression and dysthymia. The screening was facilitated
of depression among patients with diabetes in primary care.
by Group Health s prior development of a population-based
This paper describes the design of these research studies and
diabetes registry. Patients are added to the diabetes registry
the rationale for key methodologic decisions. The first part
based on: 1) currently taking any diabetic agent; or 2) a
of the Methods section will describe the recruitment for the
fasting glucose 126 confirmed by a second out-of-range
epidemiologic phase of the study, and the second part the
test within one year; or 3) a random plasma glucose 200
design of the randomized controlled trial.
also confirmed by a second test within one year; or 4) a
hospital discharge diagnosis of diabetes at any time during
GHC enrollment or two outpatient diagnoses of diabetes
2. Methods
[15]. The goal of the epidemiologic survey was to success-
fully screen at least 4,500 primary care patients with diabe-
The Pathways Study was developed by a multidisci- tes with approximately 630 expected to meet criteria for
plinary team in the Department of Psychiatry at the Univer- major depression and/or dysthymia. The goal of the ran-
sity of Washington and the Center for Health Studies at domized trial was to enroll approximately 300 depressed
patients in the randomized controlled trial. After consider-
Group Health Cooperative. Group Health is a nonprofit
ing mail versus telephone screening for depression, we
health maintenance organization with 30 primary care clin-
decided on mail screening as the most cost effective mech-
ics in Western Washington State.
anism based on prior studies by members of our study group
The study was funded by the National Institute of Mental
that were able to attain 60 to 65% recruitment rates with
Health Services (NIMH) Division of Intervention and Ser-
mail screening [7,16]. To potentially increase patient re-
vices Research. The randomized controlled trial proposed to
sponse rates, we presented the study to each of the 9 clinics
test the effectiveness of a collaborative care intervention
and requested written permission from all primary care
versus usual care. Collaborative care is a multimodal inter-
physicians to use a stamp of their signature in the approach
vention that includes integration of a  care manager (often
letter describing the study. Among the 113 doctors, 101
a nurse or mental health specialist) into primary care. The
(90%) agreed to allow us to use their signature stamp. For
 care manager works with both the patient and primary
physicians refusing to let us use their signature, we used the
care physician and helps with developing a shared definition
name of the GHC Chief of endocrinology, Dr. David Mc-
of the problem, providing patient education and support,
Culloch.
developing a shared focus on specific problems, targeting
Patients were screened by mail in sequential waves with
goals and a specific action plan, offering support and prob-
approximately 700 questionnaires sent per month. A $3 gift
lem-solving to optimize self-management, achieving closer
certificate for a local store was included with the mailing to
monitoring of adherence and outcomes, and facilitating ap-
encourage response. If the patient did not return a mailed
pointments to the primary care physician or specialist for
packet by 4 weeks, a second packet was sent. If this second
patients with adverse outcomes or side-effects [14]. The
packet was not returned by 2 weeks, the patient received a
study protocol was reviewed and approved by institutional
telephone reminder call. The first mail-screen had a re-
review boards at the University of Washington and Group
sponse rate of approximately 38%, the second mailing in-
Health Cooperative. creased the response rate to 47%, and the combination of the
telephone reminder and a last mailing 6 months after the
telephone reminder increased the final response rate to
2.1. Study setting
61.7%. Figure 1 describes the recruitment and reasons for
ineligibility or refusal at each phase of the study. The study
Nine Group Health Cooperative primary care clinics in
team has received permission from the Group Health Co-
western Washington were selected for the study. We se-
operative Institutional Review Board to collect aggregate
lected clinics based on 3 criteria: 1) clinics with the largest
data on nonrespondents to ascertain whether there are dif-
number of diabetic patients (to save screening and interven-
ferences in demographic (age, gender) or clinical variables
tion costs; 2) clinics within a 40-mile geographic radius of
(health care costs, medical comorbidity, type of diabetes or
Seattle in order to decrease travel time for nurse  care
depression treatment, and HbA1C levels) between respon-
managers ; and 3) clinics with the highest percentage of
dents and nonrespondents.
minority patients. Because minorities have higher rates of
A major methodological question was:  What is the best
diabetes, we were able to enroll substantial numbers of
depression screening tool to use? Ideally this screen should be
minority patients even though the general population was
brief, easy to score and provide both a DSM-IV diagnosis and
predominantly Caucasian. depression severity score. We elected to use the Patient Health
160 W.J. Katon et al. / General Hospital Psychiatry 25 (2003) 158  168
Fig. 1. Recruitment of epidemiologic study and randomized controlled trial.* Eligibility criteria: PHQ 10 or greater.** Patients were categorized as
 Ineligible  Other if: 1) they were enrolled in another study; 2) their spouse was enrolled in PATHWAYS; 3) they were high risk for self-harm or if they
refused a self-harm assessment; or 4) there were other special circumstances (i.e.,  there was one case where the team deemed someone ineligible due to
a recent hospitalization for drug overdose).
Questionnaire (PHQ) based on this questionnaire s ability to 2.3. Methodology of the Pathways randomized controlled
provide both a dichotomous diagnosis of major depression as trial
well as a continuous severity score [17]. The PHQ diagnosis of
major depression has been found to have high agreement with Patients were required to have a score of 10 on the
the diagnosis of major depression based on structure psychi- initial PHQ in the mail screen, which has been found to be
atric interview [17]. Because we were also interested in the the optimal cut-point in screening for major depression [17]
DSM-IV diagnosis of dysthymia, which is not included in the We required patients to have a second screen by telephone
PHQ, we added questions from the NIMH Diagnostic Inter- about two weeks after scoring 10 or greater on the PHQ. On
view Schedule [18] on dysthymia. this second screen, patients were required to have persistent
W.J. Katon et al. / General Hospital Psychiatry 25 (2003) 158  168 161
symptoms by having an SCL-20 depression [19] mean item on the intervention developed for the IMPACT Study,
score of 1.1. Double-screening eliminates those with tran- which randomized 1801 elderly primary care depressed
sient or spontaneously resolving depression. A total of 348 patients to a nurse collaborative intervention or usual care
patients were excluded based on an SCL 1.1.
[21]. A key design question for the team was:  Would this
To recruit a representative sample, we had few medical
intervention be designed to enhance treatment of depression
or psychiatric exclusions. We elected to include diabetics
only, or to improve quality of care for both depression and
who were already receiving antidepressant medication or
diabetes? We elected to design an intervention to improve
psychotherapy from nonpsychiatrist clinicians, but who still
quality of care and outcomes of depression but to not di-
had high depression scores. This decision was based on
rectly intervene to improve diabetes education or care, ex-
prior findings that showed that many primary care patients
cept to the extent that addressing diabetes care issues arose
with depression are exposed to antidepressants at lower than
in the context of treating depression. An example of a
guideline-recommended dosage and duration [20]. Eligible
diabetes issue that could have been addressed in problem
patients were ambulatory, English-speaking, with adequate
solving therapy would be if the patient chose lack of exer-
hearing to complete a telephone interview, and planned to
cise or having problems with diet as a problem she or he
continue to be enrolled in GHC over the next year. Psychi-
wanted to work on. By improving depression care, we could
atric exclusions were: 1) currently in care by, or scheduled
then test effects of improved depression outcomes on dia-
to see, a psychiatrist; 2) a diagnosis based on Group
betes self-care (diet, exercise, medication adherence), and
Health s automated diagnostic data of bipolar disorder or
glycemic control.
schizophrenia; 3) use of antipsychotic or mood stabilizer
Efficacy studies often ask questions such as  Is this
medication based on Group Health s automated pharmacy
antidepressant more effective than placebo? The health
in the prior year; and 4) mental confusion on the interview
services question for this effectiveness study was:  Is an
suggesting significant dementia (Fig. 1).
innovative method to improve service delivery that provides
guideline level antidepressant treatment or brief psychother-
2.4. Randomization
apy more effective than usual care? In developing this
intervention, we tried to optimize patient recruitment and
After completion of the baseline telephone interview and
retention by providing an initial choice based on patient
verbal informed consent, participants were informed that
preference of either antidepressant medication or problem
they would be randomly assigned to the Intervention or
solving therapy (PST). It is controversial whether providing
Usual Care group through a computer generated number.
patient choice of treatment leads to better outcomes, [22]
Patients were told that if they were assigned to the Inter-
but choice is more like  real world treatment decisions that
vention group a nurse would call them within one week to
physicians and patients negotiate. We expected choice to
set up an appointment. If they were assigned to the Usual
enhance recruitment and retention of patients. Choice of
Care group they would receive a mailed written informed
treatment is also consistent with the Institute of Medicine s
consent for telephone follow-up calls and HbA1C blood
emphasis on understanding patients beliefs and preferences
draws to sign and mail back, and the first telephone survey
in negotiating a treatment plan [23]. Problem solving ther-
call in three months.
apy was chosen because it is patient-centered, brief and
After the baseline telephone call, the research assistant
well-accepted by primary care patients due to its psycho-
handed a face sheet with a study identification number to the
educational content. Problem solving therapy has been
project coordinator to put in an Access data base. The
found to be as effective in randomized trials in primary care
Access data base then automatically generated a random
as antidepressants in improving depressive symptoms of
assignment number which indicated whether patients were
patients with major depression [24]. It was also easier to
in the Intervention or Usual Care group. Randomization
train Depression Care Specialists in providing PST than
allocation occurred in blocks of eight. For those patients in
other forms of psychotherapy.
the intervention group, the computer generated a face sheet
with a patient name and phone number that the project
2.6. What type of professional should be trained as a
coordinator delivered to the nurses. For both Intervention
Depression Care Specialist (DCS)?
and Usual Care patients the computer then added the patient
identification data to the telephone survey data base with the
Given the need for the DCS to be proficient in medica-
specific dates for the series of follow-up interviews.
tion management and PST, to have experience working with
patients with one or more chronic medical illnesses, and to
2.5. Intervention design
be comfortable working in a primary care setting, we chose
registered nurses to implement collaborative care treatment.
The intervention was an individualized, stepped care
depression treatment program provided by a Depression Registered nurses at GHC were already providing disease
Clinical Specialist (DCS) nurse in collaboration with the management for diabetes and congestive heart failure.
primary care physician. This intervention design was based Therefore, this model would have a greater chance to be
162 W.J. Katon et al. / General Hospital Psychiatry 25 (2003) 158  168
integrated into the GHC plans for improving disease man- nurse had supervision twice a month with a team of a
agement of depression after the study. psychiatrist, psychologist (on PST) and family physician to
We also required a registered nurse (R.N.) degree, not a review new cases and patient progress. Nurses interacted
nurse practitioner (ARNP), since primary care physicians regularly (via written notes and verbally) with the primary
would continue to prescribe and this level of training is care physician treating the patient. On alternative weeks,
more generalizable and cost-effective. nurses reviewed cases by telephone with the psychiatrist
We hired three half-time registered nurses. They each supervisor. The psychiatrist supervisor regularly reviewed
covered two to four primary care clinics, that were geo- choices and dosages of medication and clinical response,
graphically as far as 25 miles apart, with case loads of 40 to and recommended changes, which the nurse discussed with
65 patients each once the study was fully underway. the primary care physician and patient.
A unique clinical monitoring system was developed us-
2.7. Training ing Pendragon software [28] for a hand-held organizer for
the nurses to enter tracking data after each patient contact
Nurses received an initial one-week training course on including initial PHQ score, initial date of intake, last date
diagnosis and pharmacotherapy and an introduction to prob- seen and last PHQ score, whether the patient has had a 50%
lem solving treatment methods. A psychiatrist, primary care decrease in PHQ score by 12 weeks, initial treatment (PST
physician and psychologist participated in training. An in- or antidepressants), current treatment and number of outpa-
tervention manual from the IMPACT trial [25] was used to tient and telephone contacts. This monitoring system al-
train nurses on collaborative care, stepped care principles, lowed nurses and supervisors to easily check which patients
pharmacology and problem solving approaches. were due for telephone or in-person follow-up visits. Each
Nurses were also trained using the manual for PST-PC week these data were transferred to an Access file and an
[26] during a training period following the protocol de- updated printout of all cases was used in weekly supervi-
scribed by Hegel and colleagues [27]. Formal training in- sion. This facilitated each supervisor s review of the process
cluded didactics, role play, observation of a videotaped and outcomes of care for the large number of cases being
demonstration, and review of the treatment manual. Each managed.
nurse was required to treat at least 4 depressed patients with The printout included an asterisk for cases that had not
6 sessions of PST-PC over a 2-month period. Each session decreased 50% or more on the PHQ at 10 weeks. Supervi-
was audiotaped, and sessions 1, 3 and 5 were rated using sion started on new cases, progressed to asterisked cases and
Hegel s PST Adherence and Competency Rating Scale [27]. then to cases in initial phases of treatment.
Nurses were required to meet the criteria of at least 3 tapes
from each of two different patients audiotaped treatment 2.9. Stepped care algorithm
sessions being rated satisfactory by the team psychologist
(Dr. Ludman). During the training period, the nurses met A stepped care approach was used in which different
weekly with the psychologist for review of the audiotaped patients received different intensity of services based on
sessions. During the course of the study, the nurses met their observed outcome (Table 1). Stepped care recognizes
regularly with the psychologist to review audiotapes and that patients have marked differences in psychiatric and
specific clinical problems arising in PST sessions. Group medical comorbidity as well as differences in response to
supervision sessions were held weekly or twice a month for antidepressant medication and/or psychotherapy [29]. In the
the first months of the study, reducing in frequency over Pathways trial if patients still had persistent depressive
time. During the second year, group supervision occurred symptoms ( 50% decrease in severity based on the PHQ)
monthly. Individual PST-PC supervision sessions with 10 to12 weeks after Step 1 level treatment with either PST
nurses occurred on an as-needed basis for review of difficult or antidepressant medication, they could either: a) switch to
sessions. a second antidepressant with a different mechanism or side-
effect profile; b) switch to the alternative treatment (from
2.8. Collaborative care PST to medication or vice versa); or c) receive augmenta-
tion of PST or antidepressant medication with the first
A team of clinicians delivered the treatment for interven- treatment they had received. This change in treatment at 10
tion patients. Nurses carried out the majority of treatment to 12 weeks was labeled Step 2 care. Another option in Step
that included an initial one hour visit followed by twice a 2 was a psychiatric consultation to evaluate treatment op-
month, half-hour appointments (telephone and in-person) in tions. For patients who received one or more Step 2 inter-
the acute phase of treatment (0 to 12 weeks). The first ventions, persistent symptoms ( 50% improvement) and
appointment included a semistructured biopsychosocial his- lack of patient and clinician satisfaction with outcome after
tory, patient education, development of the therapeutic al- a second treatment (8 to 12 weeks) could lead to referral to
liance, understanding the patient explanatory model of ill- the Group Health Cooperative (GHC) mental health system
ness and negotiation whether to start treatment with an for longer term follow-up including management by a psy-
antidepressant medication or problem solving therapy. Each chiatrist (Step 3).
W.J. Katon et al. / General Hospital Psychiatry 25 (2003) 158  168 163
Table 1
Stepped care intervention
Step 1: 0 12 weeks
History taking and building therapeutic alliance
Behavior activation based on increasing positive activities
Problem-solving therapy (PST) or antidepressant medication (patient choice)
Introduction of Patient Health Questionnaire (PHQ) as weekly measure of depressive symptoms
Schedule telephone and in-person follow-ups by Depression Clinical Specialist (DCS)
Regular communication with primary care physician
,n
No Response Recovery ( 50% decrease in PHQ OR
Step 2: 12 to 24 weeks Remission (PHQ of 5)
For patients who had not decreased by at least 50% on the initial PHQ Schedule monthly continuation telephone follow-up
score and/or were dissatisfied with outcome, the DCS could:
a) switch to an alternative antidepressant medication if no or little
response to first medication
b) add an antidepressant if the patient had not responded to PST or add
PST if little or no response to antidepressant
c) augment with a second antidepressant if partial response to first
d) schedule a psychiatric consultation
,n
Less than 50% Response Based on PHQ Recovery or Remission
Step 3: 24 52 weeks Schedule monthly continuation telephone follow-up
If patient has not had decrease by 50% in PHQ or remission:
a) consider referral to Group Health mental health service for longer-
term mental health care;
b) enter patient into DCS continuation/ maintenance phase group.
Step 1 included history taking and building the therapeu- In most cases, usual care for depression provided by
tic alliance, patient choice of initial treatment and behav- Group Health Cooperative family physicians involves a
ioral activation. In addition, the depression clinical special- prescription of an antidepressant medication, 2 visits over
ist introduced the PHQ depression module as the key the first 3 months of treatment and an option to refer to
monitoring tool for measuring response to treatment and set Group Health Cooperative mental health services. Both
up a schedule of telephone and in-person sessions. Regular intervention and usual care patients could also self-refer to
communication occurred with the primary care physician a GHC mental health provider. We tracked and will report
(PCP). The goal of the intervention was clinical recovery these out-of-study mental health referrals and visits. Usual
( 50% decrease in PHQ score) and, if possible, remission care for diabetes in Group Health Cooperative is provided
(a score of 5 on PHQ) [17] and restoration of social and by the primary care physician with occasional support from
vocational function. Once patients reached a significant diabetes nurses for patients with persistently high HbA1C
decrease in clinical symptoms, the nurse began continuation levels.
phase treatment that involved monthly scheduled telephone
contacts. Nurses also set up optional continuation groups,
2.11. Evaluation
which involved a monthly group visit for patients with
persistent symptoms or social isolation instead of the
Given that patients entering the trial had evidence of at
monthly telephone calls.
least two medical illnesses, i.e., major depression and/or
dysthymia and diabetes mellitus, and that our intervention
2.10. Usual care was aimed primarily at improvement of depression, the
primary outcome variable was change in depressive symp-
Patients were informed prior to randomization that they toms. Changes in functioning were identified as important
were eligible for a new program to help people with diabe- secondary outcomes. Inherent in this discussion and the
tes better manage stress and depression. Patients were told focus of the intervention on depression was that if we
they would be called by the DCS within 10 days if they significantly improved depressive symptom and functional
were randomized to the intervention. It was also recom- outcomes, this would be considered a positive trial. We also
mended that whether or not they were chosen to receive the hypothesized that if the intervention significantly improved
additional services, they should work with their primary depressive symptoms, we would find improvement on the
care physician on these clinical issues. diabetes measures, including diabetes symptom burden, di-
164 W.J. Katon et al. / General Hospital Psychiatry 25 (2003) 158  168
abetes control and some measures of diabetes self-care who had not had one within 14 days of each study scheduled
(particularly exercise and diet). The study was adequately HbA1C.
powered to detect a moderate-to-large effect on HbA1C. We Group Health s computerized pharmacy and utilization
also planned a cost-effectiveness analysis (see below). records were used to measure adherence to antidepressant
After randomization, telephone interviews were provided medication, oral hypoglycemic medications as well as am-
at 3, 6, 12 and 24 months. Telephone interviews were bulatory visits and tests and inpatient hospital days and
completed by interviewers blind to intervention status. medical costs. The computerized pharmacy records allowed
For measuring change in depression, the SCL-20 depres- examination of refills of antidepressant medications and
sion scale [19] was chosen to be the primary dependent whether the patient received an adequate dosage based on
variable to measure change in affective symptoms based on evidence-based guideline standards for 90 days or more
previous studies showing it to be sensitive to change within each 6-month period of time. A recently developed
[30,31]. We used the PHQ at baseline, 6, 12 and 24 months algorithm for oral hypoglycemic refills also allows measure-
to measure changes in dichotomous diagnosis of major ment of whether or not the patient was overdue in refilling
depression as well as remission status (PHQ 5) [17]. his or her prescription by 15 or more days and by more than
We utilized selected scales from the WHO-DAS-II 25% of the intended duration of use [7]. Prior research has
(Household and Work-Related Activities, Community and shown that depression is associated with significant gaps in
Family Activities, Physical Health, Work Absenteeism and refills of oral hypoglycemics [7].
Cut-Down days, Work Productivity) [32] and SF-36 (Gen- Computerized pharmacy records will also be used to
eral Health, Social Role, Impairment, Emotional Role Im- compute a revised chronic disease score (Rx Risk), a mea-
pairment) [33] to measure selected domains of function. sure of chronic comorbidity based on prescription drug use
These scales were chosen because they are complementary over the previous 6 months [38]. After review of the liter-
and responsive to change in depression status. The SF-36 ature and consultation with diabetes consultants, a diabetes
does not capture all functional domains hypothesized to severity score was developed based on the patient s self-
improve with increased quality of depression care, while the rating of the number of complications of diabetes (neurop-
WHO-DAS-II has not been used in prior randomized trials. athy, nephropathy, retinopathy and myocardial infarction)
We utilized well-validated, reliable scales to measure [39] the patient has had, whether the patient s initial diabe-
diabetes symptom burden (9 items) [34], diabetes self-effi- tes medication was insulin or an oral hypoglycemic, and the
cacy (7 items) [35], and diabetes self-care activities (12 length of time he or she had diabetes. Both the chronic
items) [36]. We selected these three scales because they disease score and diabetes severity score will be used as
were shown to have a high correlation with severity of covariates in assessing intervention versus usual care out-
depression in a prior epidemiologic study [7]. The Diabetes comes. Patients were defined as having Type 1 diabetes
Symptom Burden Scale inquires about 9 symptoms in the based on an age of onset less than age 30 and on insulin
last month, such as abnormal thirst, and codes the answer on being their first and current treatment.
a Likert Scale of 1 (never) to 5 (qd) [34]. The Diabetes Computerized health plan data will be used to identify all
Self-Efficacy Scale inquires about 7 items, such as how health plan services provided or paid for by Group Health
much control over diabetes the patient believes she or he Cooperative during the 12- and 24-month periods after
has, on a 1 (none at all) to 5 (total control) Likert Scale [35]. randomization (inpatient and outpatient services for mental
The Diabetes Self-Care Activity Scale asks patients to eval- health or general medical care). All outpatient and inpatient
uate how many of the last 7 days they have followed an services provided by Group Health Cooperative are as-
exercise or diet program, checked on their blood sugar level signed costs based on health plan accounting records (in-
and completed a foot check. Each answer is coded on a 0 cluding actual personnel, supply and overhead costs). Ser-
(no days) to 7 (days) Likert Scale [36]. vices purchased by GHC from external providers are
Enrolled patients were asked to agree to blood draws to assigned costs equal to the amount reimbursed by Group
measure HbA1C at baseline, 6,12 and 24 months and were Health Cooperative for that type of care.
reimbursed $25 for their time for each test. Virtually all
enrolled patients were willing to participate in the blood 2.12. Data analysis
draws. HbA1C measures exposure of red blood cells to
glucose over a 120-day period, and diabetes guidelines The study was powered using three different hypothe-
recommend that primary care physicians order this test sized differences between Intervention and Usual Care pa-
twice a year [3]. Lowering HbA1C levels has been targeted tients including depressive symptoms, function and HbA1C
in patients with diabetes as a key mechanism to decrease levels. Based on previous studies in which interventions and
medical sequelae of poor diabetes control. The Diabetes controls had a 0.31 difference on the SCL-20 at 6 months
Complications and Control Trial demonstrated that inter- with a standard deviation for the SCL-20 of 0.7 (an effect
ventions that significantly lowered HbA1C levels in patients size of approximately 0.4), the study design had 80% power
with diabetes decreased important medical complications to detect a 0.23 difference between groups, assuming ran-
[37]. Operationally, we required a new HbA1C on patients domization of 300 patients and 85% patient retention in the
W.J. Katon et al. / General Hospital Psychiatry 25 (2003) 158  168 165
study at 12 months. Based on previous primary care depres- Methods to correct for possible heteroscedasticity in cost
sion studies we expected both the SF-36 Social Role Func- data such as  smearing techniques [42] or gamma regres-
tioning and Emotional Role Functioning Scales to have a sion may be utilized [43]. The randomized trial will have
standard deviation of 30. Based on the sample size of 300 adequate power to detect a $1,000 reduction in ambulatory
and 85% retention, we calculated that we would have 80% costs per year from a projected base of $5,000 per year. This
power to detect a 10-point difference in SF-Social or Emo- reduction is greater than what we expect to observe in the
tion Role Function scores, which is considered clinically intervention group. However, descriptive trends in health
significant. Based on a recent study of depression and dia- care costs in intervention and control groups may be im-
betes in primary care [7] we assumed a standard deviation portant. If depression and/or diabetes outcomes were im-
of HbA1C levels of 1.65 and a correlation between baseline proved without evidence of increased health care costs be-
and follow-up levels of 0.6 (our actual estimates are 0.7). tween intervention and usual care groups, this would be of
With a sample size of 145 in each group we will have 82.2% interest. We plan to attempt to obtain a better understanding
power to detect a 0.5 difference between Intervention and of the effect of the interventions on inpatient hospitalization
Controls. Given that we expect an 85% follow-up rate at 12 in two analyses. After controlling for age, gender, chronic
months, we projected needing 162 per group or a total disease score and diabetes severity, we will examine the
sample size of 324 to be able to detect a HbA1C difference effect of the intervention on the number of hospitalizations
of 0.5. using a negative binominal regression model. We will also
The analyses of outcome differences between interven- examine the effect of the intervention on time to first hos-
tion and control patients will follow an intent-to-treat ap- pitalization in the first year period using survival models.
proach. We will use random regression models to estimate Costs and effectiveness of the intervention will be com-
effects of the intervention relative to usual care, on the pared with incremental cost-effectiveness ratios following
primary dependent variables: depressive symptoms (contin- guidelines developed by Gold and colleagues [44]. We will
uous SCL-20 outcomes as well as percent achieving recov- take a societal perspective and, in the numerator, will esti-
ery based on 50% decrease on SCL-20 and the patient mate the one-year differences in total ambulatory costs and
achieving remission based on a score of 5 on PHQ or time off work due to medical or mental health visits. In the
0.5 on SCL) and function (WHO-DAS and SF-36 sub- denominator we will use the method by Lave et al. [45] to
scales). We will test two interaction terms in the model to estimate differences in depression free days between inter-
examine possible differences in intervention effects: out- vention and control patients over a 12-month period. Boot-
comes for insulin-dependent versus noninsulin-dependent strap resampling with 1,000 draws using bias correction will
diabetics and outcomes for patients with HbA1C levels less be used to estimate confidence intervals for both incremen-
than and greater than guideline recommended levels. Pre- tal cost measures and depression-free days and the ratio of
vious studies suggest that depression may have a greater incremental costs to incremental depression-free days [46].
impact on HbA1C levels in insulin-dependent diabetic pa- The bootstrap method will allow us to document the prob-
tients [40] and change in depression should have the most ability of this intervention being in each of the four quad-
benefit in patients with high HbA1C levels. We will also use rants in Fig. 2. Most new interventions are in the upper left
random regression models to compare intervention versus quadrant (costs more, but more effective), however, there is
usual care effects on important secondary outcomes: diabe- also a possibility that, if improved depression care is asso-
tes symptom burden, diabetes self-efficacy, diabetes self- ciated with improved diabetes care, there may be savings in
care, and HbA1C. In these models, we will adjust for dif- medical costs that partially or completely make-up for in-
ferences in patient demographics and clinical characteristics creased depression costs inherent in the collaborative care
across intervention and control patients. model. Because depression costs are mostly increased in the
first 6 months and medical cost savings may be delayed, we
2.13. Health care costs and cost effectiveness will carry out follow-ups over a 2-year period.
We will follow patient costs over a 2-year period after
randomization (the initial patient was randomized on 4/27/ 3. Results
2001 and the last patient was randomized on 5/8/02), Pre-
vious data on health care costs in patients with depression Upon completing enrollment, 330 primary care patients
and diabetes for a 6-month period were estimated at $3,654 with diabetes were randomized to the intervention or usual
$4,258 in patients with depression and $2,094 $3,052 care conditions. Table 2 presents demographic and clinical
in patients without depression [7]. Methodologic issues ad- characteristics of the randomized subjects. Approximately
dressed by the skewed distribution of heath care costs in- 23% of subjects are of minority ethnicity which is higher
clude: using two-part models [41] which first uses logistic that the rates in the Group Health system. This is because of
regression to compare the percent of patients utilizing any selecting clinics with high minority rates and the higher
health care services. Linear regression techniques are then percentage of non-Caucasians with diabetes. A substantial
used to compare health care costs among users of services. number of patients have comorbid medical disorders in
166 W.J. Katon et al. / General Hospital Psychiatry 25 (2003) 158  168
Fig. 2. Incremental cost effectiveness quandrant
addition to diabetes and are from low socioeconomic status 4. Discussion
both of which have been found to be risk factors for poor
depression outcomes [47,48]. The Pathways project has demonstrated the feasibility of
The importance of choice in treatment was shown by the recruiting a population-based primary care sample of pa-
exposure to prior treatment and the actual choices patients tients with diabetes and depression for an epidemiologic
initially made for treatment. Of the 165 patients randomized study and a randomized controlled trial. The mail survey
to the collaborative care intervention, 48 (29.0%) initiated
coupled with telephone reminder calls successfully screened
treatment with PST only, 65 (39.4%) initiated treatment
61.7% of the population. We will be able to compare non-
with medication only, 48 (29.0%) initiated treatment with
respondents to respondents on multiple variables in the
PST and medication (because they were already on an
GHC database (i.e., prior utilization and costs, medical
antidepressant, but still had significant depressive symp- comorbidity, HbA1C levels) to ascertain respondent bias.
toms based on a screening PHQ score of 10 and baseline
The Pathways intervention offers patients and providers
SCL of 1.1), and only 4 (2.4%) never initiated treatment.
the necessary resources to increase the use of evidence-
based depression treatments. The nurse collaborative care
model exemplifies a system of care that both supports the
Table 2
primary care delivery system and provides patient-centered
Demographics and clinical characteristics of patients with diabetes and
care. This intervention was modeled from the IMPACT
depression
study where patients in the intervention arm were signifi-
Intervention Control
cantly more satisfied with care over the first 3 months than
N 165 N 165
those treated in usual care [49]. The provision of choice of
Age (mean SD) 58.6 11.8 58.1 12.0
treatments may have helped with speed of recruitment and
% Female 64.7% 64.7%
retention. Many patients had negative feelings about one of
% 1 year of college 80.0% 77.6%
the two treatments. For instance, some patient stated they
% White 75.3% 81.1%
% Employed full- or part-time 53.9% 45.2%
were already on multiple medications and wouldn t take
SCL-depression (mean SD) 34.15 10.2 32.5 9.1
another, whereas others were not interested in counseling
% Lifetime dysthymia 67.7% 70.3%
but agreed to try a medication. Choice mirrors how patients
% Major depression 62.8% 69.1%
and providers work together in  real world systems and
% Current panic disorder 9.6% 11.9%
Years with diabetes (mean SD) 9.6 8.7 10.2 10.1 should provide more realistic estimates of the feasibility and
Total diabetes symptoms (0 10) 4.6 2.6 4.7 2.3
effectiveness of such treatments in primary care.
(mean SD)
The stepped care model is more complex than most
HbA1C (mean SD) 8.1 1.6 8.0 1.5
treatment protocols and targets scarce mental health re-
W.J. Katon et al. / General Hospital Psychiatry 25 (2003) 158  168 167
[15] McCulloch D, Price M, Hindmarsh M, Wagner E. A population-
sources to patients with the most persistent symptoms. This
based approach to diabetes management in a primary care setting:
is not a trial of PST versus medication versus usual care;
early results and lessons learned. Eff Clin Pract 1998;1:12 22.
instead it is a trial of a health services intervention that
[16] Walker E, Unutzer J, Rutter C, et al. Costs of health care use by
provides a choice of evidence-based depression treatments
women HMO members with a history of childhood abuse and neglect.
versus usual care. We will not be able to analyze which
Arch Gen Psychiatry 1999;56:609  13.
[17] Kroenke K, Spitzer RL, Williams JB. The PHQ-9: validity of a brief
components of this multifaceted intervention are most im-
depression severity measure. J Gen Intern Med 2001;16:606 13.
portant in improving outcomes; however, reviews of
[18] Robins L, Helzer J. Diagnostic interview schedule (DIS): version
chronic disease interventions that have successfully im-
III-A. St Louis, MO, Washington University School of Medicine,
proved patient-level outcomes have shown that interven-
1985.
tions aimed at multiple levels of care, including the patient,
[19] Derogatis L, Rickels K, Uhlenhuth E, Coui C. The Hopkins Symptom
Checklist: a measure of primary symptom dimensions. In: Pichot
physician and process of care are most effective [50].
(editor), Psychological measurement in psychopharmacology: prob-
lems in psychopharmacology. Basel, Switzerland: Kargerman, 1974,
p. 79 110.
Acknowledgments
[20] Simon G. Evidence review: efficacy and effectiveness of antidepres-
sant treatment in primary care. Gen Hosp Psychiatry 2002;24:213 24.
Supported by grants #MH 4 1739 and #MH 016473 [21] Unützer J, Katon W, Williams J Jr, et al. Improving primary care for
depression in late life: the design of a multicenter randomized trial.
from the National Institute of Mental Health Services Di-
Med Care 2001;39:785 99.
vision, Bethesda, MD (Dr. Katon).
[22] Ward E, King M, Lloyd M, et al. Randomized trial of nondirected
counseling, cognitive-behavioral therapy, and usual general practitio-
ner care for patients with depression. clinical effectiveness. BMJ
References
2000;321:1383 88.
[23] Institute of Medicine. Crossing the quality chasm: a new health
system for the 21st century. Washington, D.C., National Academy
[1] Chipkin SR, Gottlieb P, Bogorad DD, Parker F. Diabetes mellitus. In:
Press, 2001.
Noble J, editor. Textbook of primary care medicine. 2nd ed. St. Louis,
[24] Mynors-Wallis L, Gath D, Lloyd-Thomas A, Tomlinson D. Random-
Mosby-Yearbook, 1996, p. 476 498.
ized controlled trial comparing problem solving treatment with ami-
[2] Lustman P, Clouse R, Freedland K. Management of major depression
triptyline and placebo for major depression in primary care. BMJ
in adults with diabetes: implications of recent trials. Sem Clin Neu-
1995;310:441 5.
ropsych 1998;3:102 114.
[3] Centers for Disease Control and Prevention. National diabetes aware- [25] Unützer J. The IMPACT study investigators: IMPACT intervention
manual. Los Angeles, CA, Center for Health Services Research,
ness month, Nov 1997, MMWR, Morb Mortal Wkly Rep 46:1013,
UCLA Neuropsychiatric Institute, 1999.
1997.
[26] Hegel MT, Barrett JE, Oxman TE Problem-solving treatment for
[4] American Diabetes Association, Direct and indirect costs of diabetes
primary care (PST-PC): A treatment manual for depression, Hanover
in the United States in 1992, Alexandria, VA, 1992.
NH: Dartmouth University, 1999.
[5] Gavard JA, Lustman PJ, Clouse PE. Prevalence of depression in
[27] Hegel MT, Barrett JE, Oxman TE. Training United States therapists
adults with diabetes. An epidemiologic evaluation. Diabetes Care
in problem-solving treatment of depressive disorders in primary care
1992;16:1167 78.
(PST-PC): lessons learned from the treatment effectiveness project.
[6] de Groot M, Anderson R, Freedland K, et al. Association of depres-
Families Systems Health 2000;18(4):423 35.
sion and diabetes complications: meta-analysis. Psychosom Med
[28] Pendragon Forms, Version 3.2. Pendragon Software Corporation,
2001;63:619 30.
Libertyville, IL 60048. www.Pendragonsoftware.com.
[7] Anderson R, Freedland K, Clouse R, Lustman P. Prevalence of
comorbid depression in adults with diabetes. A meta-analysis. Dia- [29] Katon W, Von Korff M, Lin E, Simon G. Rethinking practitioner
betes Care 2001;24:1069 78. roles in chronic illness: the specialist, primary care physician, and the
[8] Lustman PJ, Anderson R, Freedland K, et al. Depression and poor practice nurse. Gen Hosp Psychiatry 2001;23:138 44.
glycemic control: a meta-analytic review of the literature. Diabetes [30] Katon W, Von Koff M, Lin E, et al. Collaborative management to
Care 2000;23:934 42. achieve treatment guidelines: impact on depression in primary care.
[9] Winokur A, Maislin G, Phillips J, Amsterdam J. Insulin resistance JAMA 1995;273:1026 31.
after glucose tolerance testing in patients with major depression. Am J [31] Katon W, Robinson P, Von Korff M, et al. A multi-faceted interven-
Psychiatry 1988;145:325 30. tion to improve treatment of depression in primary care. Arch Gen
[10] DiMatteo MR, Lepper HS, Croghan TW. Depression is a risk factor Psychiatry 1996;53:924 32.
for noncompliance with medical treatment: Meta-analysis of the ef- [32] Epping-Jordan J, Chatterji S, Ustun T. The World Health Organiza-
fects of anxiety and depression on patient adherence. Arch Intern Med tion. Disability assessment schedule II (WHO-DAS II) a tool for
2000;160:2101 7. measuring clinical outcomes. Presented at NIMH Health Services
[11] Lustman PJ, Griffith LS, Clouse RE, et al. Effects of nortriptyline on Research Meeting. Washington, D.C., July 2000.
depression and glucose utilization in diabetes: results of a double- [33] Ware J, Sherbourne C. The MOS 36-item Short Form Health Survey
blind placebo controlled trial. Psychosom Med 1997;59:241 50. (SF-36). Med Care 1992;30:473 83.
[12] Lustman PJ, Freedland KE, Griffith LS, Clouse RE. Fluoxetine for [34] Whitty P, Steen N, Eccles M, et al. A new completion outcome
depression in diabetes: a randomized double-blind placebo-controlled measure for diabetes: is it responsive to change? Qual Life Res
trial. Diabetes Care 2000;23:618 23. 1997;6:407 13.
[13] Lustman PJ, Griffith LS, Freedland KE, et al. Cognitive behavior [35] Carey MP, Jorgenson RS, Weinstock, et al. Reliability and validity of
therapy for depression in type 2 diabetes mellitus: a randomized the appraisal of diabetes scale. J Behav Med 1991;14:43-51.
controlled trial. Ann Intern Med 1998;129:613 21. [36] Toobert D, Hampson S, Glascow R. The summary of the Diabetes
[14] Von Korff M, Gruman J, Schaefer J, et al. Collaborative management Self-Care Activities Measure: results from 7 studies and a revised
of chronic illness. Ann Intern Med 1997;127:1097 1102. scale. Diabetes Care 2000;23:943 50.
168 W.J. Katon et al. / General Hospital Psychiatry 25 (2003) 158  168
[37] The DCCT Research Group. Influence of intensive diabetes treatment [45] Lave J, Frank R, Schulberg H, Kamlet M. Cost-effectiveness of
on quality-of-life outcomes in the Diabetes Control and Complica- treatments for major depression in primary care practice. Arch Gen
tions Trial. Diabetes Care 1996;19:195 203. Psychiatry 1998;55:645 51.
[38] Fishman P, Goodman M, Hornbrook M. Risk adjustment using auto- [46] O Brien B, Drummond M, Labelle R, Williams A. In search of
mated pharmacy data: the Rx Risk Model. Med Care 2003;41:84 99. power and significance: issues in the design and analysis of sto-
[39] Jacobson A, de Grout M, Samson J. The effects of psychiatric dis- chastic cost-effectiveness studies in health care. Med Care 1994;
orders and symptoms on quality of life in patients with type 1 and 32:150  63.
type 2 diabetes mellitus. Qual Life Res 1997;6:11 20. [47] Weich S, Churchill R, Lewis G, et al. Do socioeconomic factors
[40] Ciechanowski P, Katon W, Russo J, Hirsch I. The relationship of predict the incidence and maintenance of psychiatric disorders in
depressive symptoms to symptom reporting, self-care and glucose primary care. Psychol Med 1997;27:73 80.
control in diabetes. Gen Hosp Psychatry (In Press). [48] Cole M, Bellavance F, Mansour A. Prognosis of depression in elderly
[41] Diehr R, Yanez D, Ash A, et al. Methods for analyzing health care community and primary care populations: a systematic review and
utilization and costs. Ann Rev Public Health 1999;20:125 144. meta-analysis. Am J Psychiatry 1999;156:1182 9.
[42] Duan N. Smearing estimate: a nonparametric retransformation [49] Unützer J, Katon W, Williams J Jr, et al. The IMPACT trial: collab-
method. J Am Statistical Assoc 1983;78:605 10. orative care management improves treatment and outcomes of late-
[43] Manning WG. The logged dependent variable, heteroscedasticity, and life depression. JAMA 2002;288:2836 45.
the retransformation problem. J Health Econ 1998;17:283 95. [50] Haynes R, McDonald H, Garg A, Montague P. Interventions for
[44] Gold M, Siegel J, Russel L, et al. Cost-Effectiveness in Health and helping patients to follow prescriptions for medications. Cochrane
Medicine. New York, NY, Oxford University Press, 1996. Database Syst Rev 2002;(2):CD00011.