Abstract Session E1: Organization of Care and Chronic Disease - - PDF document

Abstract Session E1: Organization of Care and Chronic Disease Management

Moderator: Nancy A. Rigotti, MD

UNDERSTANDING VARIATION IN PCP REFERRAL PATTERNS IN A LARGE MULTISPECIALTY PRACTICE GROUP Michael L. Barnett1; Thomas D. Sequist1,2. 1Brigham and Women's Hospital, Boston, MA; 2Partners Healthcare System, Boston, MA. (Tracking ID #1937624) BACKGROUND: Primary care physicians (PCPs) generate the vast majority of specialty referrals, and the decision to refer directly impacts health care quality, patient experiences, and spending. We analyzed referral rates among PCPs to characterize the relative contribution of patient and physician characteristics to the probability of referral, as well as the long term impact on utilization among PCPs with high rates of referral. METHODS: We analyzed electronic health record data within a large multispecialty group practice that requires electronic referral

• rders. We enrolled 78,485 patients 18 years and older who visited 142 PCPs during a baseline referral rate measurement period

(2005-2006), and then analyzed subsequent specialist referrals among these patients from 2007-2011. We collected information on patient age, gender, race, Charlson comorbidity score (2005-2006), and number of subsequent specialist appointments (2007-2011). To calculate PCP referral rates, we estimated a mixed effects logistic regression model using 2005-2006 data adjusted for patient characteristics, using a random intercept term to account for correlation within individual PCPs. We calculated the ratio of the PCP's adjusted referral rate estimated from the fitted random intercept over the expected referral rate estimated with the same model without the random intercept. We multiplied this ratio by the average referral rate in the entire cohort to calculate the case-mix adjusted referral rate for each PCP. We used the c-statistic to assess the relative contribution of patient and physician characteristics in this

• model. We next categorized PCPs according to their quintile of adjusted referral rate and examined the characteristics of the PCPs and

their patient panels, testing for trends using logistic regression. In the period from 2007-2011, we estimated the adjusted probability with logistic and negative binomial models that patients with PCPs in the different referral quintiles received any specialty referral, the average number of referrals per patient, and the average number of specialty visits. RESULTS: From 2007-2011, the PCPs placed 102,276 referral orders, representing 74% of the total referrals for this cohort (additional referrals came from urgent care physicians). The five most commonly referred specialties were orthopedic surgery (23%), dermatology (15%), otorhinolaryngology (9%), gastroenterology (9%), and general surgery (8%). Physicians with more female, non- white and younger patients as well as those with fewer comorbidities were more likely to be in the highest referral quintile, as were physicians with fewer patient encounters and smaller panels (all p<0.001, Table). The c-statistic for a logistic regression model to predict referrals using patient characteristics alone was 0.56, while the mixed effects model with a physician-level random intercept alone had a c-statistic of 0.68, which was unchanged after incorporating patient characteristics into the model. The adjusted average rate of referral per 100 patient visits was 5.2, 16.3, 21.0, 25.6, and 36.2 by quintile of referral rate among PCPs, with an overall average of 17.4. Adjusting for age, sex, race, and comorbidities, the average patient with a PCP in the highest quintile of referral rate had a 77% (95% CI 76-78) probability of receiving a referral and had 2.2 (95% CI 2.1-2.3) referrals on average from 2007-2011, compared to a 36% (95% CI 35-37) chance of receiving a referral and 0.59 (95% CI 0.57-0.61) average number of referrals in lowest

• quintile. From 2007-2011, in the 12 months subsequent to a referral, patients of PCPs in the highest referral quartile experienced an

average of 3.6 (95% CI 3.4-3.8) follow-up specialist visits versus 0.53 (95% CI 0.51-0.56) visits among patients of PCPs in the lowest referral quartile. CONCLUSIONS: We observe wide variation among PCPs in specialty referral rates, which is explained in large part by physician characteristics as opposed to patient characteristics. This variation has substantial long-term implications, with patients seen by PCPs in the highest quintile of referral rates experiencing dramatically more specialist visits over time. Our analyses suggest that physician- level interventions are needed to address this variation which may be due to physician subjectivity in the decision to refer patients. Table: Patient and Physician Characteristics, by Referral Quintile

MINDFULNESS BASED COGNITIVE THERAPY VERSUS A HEALTH ENHANCEMENT PROGRAM FOR TREATMENT RESISTANT DEPRESSION: A RANDOMIZED CONTROLLED TRIAL Mitchell D. Feldman1; Erin P. Gillung2; Kevin Delucchi3; Stuart J. Eisendrath4. 1UCSF, San Francisco, CA; 2UCSF, San Francisco, CA; 3UCSF, San Francisco, CA; 4UCSF, San Francisco, CA. (Tracking ID #1927292) BACKGROUND: Major depressive disorder (MDD) is the leading cause of disability in the developed world, yet broadly effective treatments remain elusive. Up to 40% of patients are unresponsive to at least two trials of antidepressant medication and are thus labeled as having treatment- resistant depression (TRD). There is an urgent need for cost-effective, non-pharmacologic, evidence-based treatments for TRD. Prior research has demonstrated that Mindfulness-Based Cognitive Therapy (MBCT) is an effective treatment for major depression, but it has not been previously studied in patients with TRD. MBCT is based on a combination of mindfulness meditation with elements of cognitive behavior therapy. The purpose

• f this study was to evaluate whether (MBCT) is an effective augmentation of antidepressants for adults with MDD who failed to respond to standard

pharmacotherapy. METHODS: Randomized controlled trial of MBCT versus an active comparator condition, the Health-Enhancement Program (HEP), comprised of physical fitness, nutrition and music therapy. Participants were age 18 years and older with TRD who had failed to respond to two or more antidepressant trials. All participants were taking antidepressants at the time of enrollment. One hundred seventy three participants were recruited from primary care and other settings and randomly assigned to 8 weekly group sessions of MBCT or HEP. Treatment response and depression remission rates were assessed at weeks 4 and follow-up weeks 8, 24, 36 and 52 using the clinician-rated Hamilton Depression Severity Rating Scale (HDRS). HDRS response and remission rates and mean HDRS total scores were compared between treatment conditions using a GEE-based repeated measures model accounting for clustering by cohort. The models included treatment condition, assessment point, and their interaction. RESULTS: Significant improvement was seen in rates of response (p=<.001), remission (p<.01), severity (p<.001) and percent reduction in severity score (p<.01). No significant differences between treatment groups were found. A significant condition-x-time interaction was observed for both the severity score and percent reduction indicating that the MBCT continued to improve at Week 8 while the improvement in the HEP condition leveled

• ff.

CONCLUSIONS: Both MBCT and HEP produced improvement in patients with treatment-resistant depression over 8 weeks. While the differences between conditions were not statistically significant, differences in course of improvement suggests differences in long-term follow-ups (underway) may be significant. Clinical Characteristics of Adults with Treatment-Resistant Depression Receiving Mindfulness-Based Cognitive Therapy (MBCT) or the Health Enhancement Program (HEP)

MBCT (N=87) HEP (N=86) Variable Mean SD Mean SD Age at depression onset 18.8 10.9 3.5 13.2 Total # depressive episodes (months) 3.6 2.6 78.5 2.4 Length of current depressive episode (months) 84.4 119.5 17.4 93.5 HAM-D Score 18.3 3.4 3.5 Single episode (%) 20.7 22.1 ≥3 lifetime episodes 62.2 58.0 Previous treatment for depression (%) Hospitalization 16.1 18.6 Suicide Attempt 19.0 20.5 Recruitment Source % GIM 34.5 36.1 Psychiatry Clinic 43.7 39.5 Community 24.4 24.4

HAM-D = Hamilton Depression Rating Scale Depression Outcomes over Time for Adults with Treatment-Resistant Depression Receiving Mindfulness-Based Cognitive Therapy (MBCT) or the Health Enhancement Program (HEP)

MBCT Group (N=87) HEP Group (N=86) Outcome Variable Mean ± SD Mean ± SD HAM-D Mean Score Baseline 18.3 ± 3.4 17.4 ± 3.5 Week 4 13.8 ± 4.7 12.8 ± 4.3 Week 8 11.4 ± 4.9 12.5 ± 5.0 HAM-D Percent Reduction Week 4 0.23 ± 0.20 0 .24 ± 0.24 Week 8 0.36 ± .25 0.25 ± 0.27

HAM-D = Hamilton Depression Rating Scale. Reduction rates were calculated as mean percent change from baseline to weeks 4 and 8.

A RANDOMIZED TRIAL OF A COMMUNITY HEALTH WORKER LED INTERVENTION TO IMPROVE DIABETES INTERMEDIATE OUTCOMES AMONG LATINOS PATIENTS WITH POORLY CONTROLLED DIABETES. Olveen Carrasquillo; Yisel Alonzo; Cynthia Lebron; Natalie Ferras; Ernesto Reyes-Arrechea; Hua Li; Sonjia

• Kenya. University of Miami, Miami, FL. (Tracking ID #1936786)

BACKGROUND: Diabetes disproportionately impacts Latinos. Community Health Workers (CHWSs) are one approach that may improve health outcomes in this population. However, evidence from rigorous randomized studies is limited. METHODS: The Miami Health Heart Initiative examined the impact of a comprehensive community health worker (CHW) intervention on diabetes intermediate outcomes of systolic blood pressure (SBP), lipids (LDL) and Hemoglobin A1C (HbA1c). We recruited 300 Latino patients with poorly controlled diabetes (HbA1c >=8.0%) ages 35-65 from the primary care clinics of Miami-Dade county's public hospital. Subjects were randomized to usual health care or a comprehensive structured CHW one year intervention consisting of home visits, phone calls and group education sessions. The intervention included patient navigation as well as assistance with social and non-medical needs. An RA blinded to group assignment conducted initial and follow- up evaluations at 12 months. We used linear mixed models to statistically test for the impact of our intervention

• n outcomes using intention to treat analyses.

RESULTS: The mean age of our patients was 56 + 7 years, 55% were female and mean BMI was 31.6 + 7.4 kg/m2. Over half (55%) were sedentary (IPAQ) and median daily fruit and vegetable was 2 (BRFSS). Cubans made up 29% of our sample with no other Hispanic ethnic subgroup representing over 15% of the sample; 47%

• f respondents scored the lowest possible acculturation score (Marin-Marin). Intervention patients received a

median of 5 home CHW visits and 25 phone calls; 84% of intervention group participants had at least 12 CHW contacts over the course of the year. Participation at group session was more skewed with 52% of participants not attending any sessions. Of those that attended at least one session, average attendance was 4 sessions per

• participant. At one year the retention rate was 72% and similar in both arms. Data on outcomes by group is

shown in Table 1. CONCLUSIONS: In a heterogeneous Latino population with poorly controlled diabetes, a rigorous CHW intervention resulted in statistically and clinically meaningful changes in HbAIC but not in blood pressure or

• lipids. The impact of the intervention on the latter two outcomes may have been limited by the fact nearly half
• f patients were at the SBP and LDL targets prior to the intervention.

Figure 1: Baseline and One Year Outcomes in MHHI Usual Care CHW HbA1C (%)* Baseline 9.03 9.19 One Year 9.25 8.81 Change + 0.21

• 0.38

SBP (mm/Hg) Baseline 135 131 One Year 134 128 Change

• 1
• 3

LDL (mg/dl) Baseline 108 100 One Year 111 99 Change + 3

• 1

* p= 0.01 for HbA1C NS for SBP and LDL

WEATHERING THE STORM: THE IMPACT OF HURRICANE SANDY ON PRIMARY CARE PRACTICE AT THE NEW YORK VA MEDICAL CENTER Mark D. Schwartz1,3; Ashley E. Jensen2,1; Matthew Beyrouty2,1; Katelyn Bennett1,2; Scott Sherman1,3; Joseph Leung1,2; Neil Shapiro1,2.

1VA NY Harbor Healthcare System, New York, NY; 2NYU School of Medicine, New York, NY; 3NYU School of Medicine, New

York, NY. (Tracking ID #1937759) BACKGROUND: Superstorm Sandy dramatically disrupted primary care (PC) services at the Manhattan campus of the VA New York Harbor Healthcare System (NY Harbor). The hospital was evacuated and closed on October 28, 2012, and remained closed for 6 months to undergo major repairs. Previous research of the impact of natural disasters on PC utilization is limited, but Hurricane Katrina in New Orleans highlighted the value of quickly establishing flexible "medical homes" following a disaster. In 2010 the VA established its own medical home model (Patient Aligned Care Teams, PACT), which includes strategies for panel management and expanding health care beyond the in-person encounter. During 6 months of displacement, NY Harbor staff provided PC services remotely by telephone, secure messaging, and walk-in access clinics at other VA facilities throughout New York City. The storm abruptly accelerated the need for panel management and remote care strategies, creating a unique opportunity to study how PC staff and patients coped with these changes. METHODS: In this observational study, 48 PC staff that care for approximately 16,000 veterans were surveyed 2 months following the storm (Displacement), and 6 months after returning to the hospital (Return) about the impact of the storm on their practice, stress, satisfaction, and perceived impact on patient care. Staff responded to a 7-item measure of panel management self-efficacy (11-point scale, Cronbach's α=0.9, maximum score of 70 ), and a 3-item scale of confidence in providing remote care (11-point scale, α=0.8, maximum score of 30. Using VA administrative data, PC service use including visits, telephone encounters, and secure messages was tracked to compare utilization patterns before the storm, during Displacement, and after Return. We also compared the proportions of each providers' hypertensive patients with blood pressure (BP) >140/90, and diabetic patients with hemoglobin A1c >9.0% before and after the storm. We tested the correlation between clinicians' panel management self-efficacy and the proportions of their patients with uncontrolled BP and A1c values. RESULTS: Of the 36 respondents at Displacement (77% response rate), 27% had to evacuate their homes during the storm, 72% lost power for >24 hours, and 31% reported damage to their property. During Displacement, 35% felt that patients were well cared for, but upon Return this increased to 78% (p<0.001). During Displacement, 77% said that when they return to the hospital, patients will need to be seen in person less often, 67% would provide more care by telephone, and 52% would use more secure messaging. Upon Return,

• nly 48% of staff reported that they believed patients need to be seen in person less often (p=0.04), however 70% said they provide

more care by phone and 67% said they use more secure messaging than before the storm. From Displacement to Return, the proportion with greater work stress (compared to before the storm) decreased from 51% to 45% (p=0.80), while job satisfaction increased from 45% to 75% (p=0.05). The staffs' panel management self-efficacy increased from a mean (SD) of 34.9 (10.9) during Displacement to 44.0 (15.1) on Return (26% increase, p=0.002). Staff confidence in providing remote care also trended 8.5% higher (p=0.18). Provider's with higher self-efficacy scores for panel management had fewer diabetic patients with out of range A1c readings for the 3 months immediately following the storm (r=-0.47, p=0.03) and 6 months following the storm (r=-0.52, p=0.02), but not prior to the storm; while the proportion of hypertensive patients with out of control BP was not correlated with panel management self- efficacy (r=0.36, p=0.59). CONCLUSIONS: Following Hurricane Sandy, PC staff at the NY Harbor were forced to innovatively accelerate the deployment of panel management and remote care strategies, through which they successfully maintained access to care for their patients. This experience has had an impact on the PACT PC model, with substantially increased self-efficacy in panel management (linked to better diabetes control), and improved confidence in providing remote care. Growth in remote care was sustained upon return to the hospital. These results argue for expanding the training of medical home teams in panel management and remote care strategies. These findings thus have implications for disaster planning as well as for advancing medical home models. Average Monthly Primary Care (PC) Encounters and Outcomes Before and After Hurricane Sandy Before Storm (May-Oct. 2012) Displacement (Nov. 2012-Apr. 2013) Return (May-Oct. 2013) Total # of PC of Encounters 6,356 5,251 5,643 % of encounters by phone 25.2 45.5 28.4 Total # of secure messages 350 561 577 % of hypertensive patients with BP>140/90 23.6 28.1 28.3 % of diabetic patients with A1c>9% 23.8 25.1 28.2

THE USE OF PANEL MANAGEMENT ASSISTANTS TO IMPROVE SMOKING CESSATION AND HYPERTENSION MANAGEMENT BY VA PRIMARY CARE TEAMS: A CLUSTER RANDOMIZED CONTROLLED TRIAL Mark D. Schwartz1,2; Ashley E. Jensen1,2; Binhuan Wang1,2; Katelyn Bennett1,2; Anne Dembitzer1,2; Shiela Strauss3; Antoinette Schoenthaler2; Colleen Gillespie2; Scott Sherman1,2. 1VA New York Harbor Healthcare System, New York, NY; 2NYU School of Medicine, New York, NY; 3NYU College of Nursing, New York,

• NY. (Tracking ID #1937793)

BACKGROUND: Panel Management (PM), a set of tools and processes applied to populations of patients, can expand prevention and chronic illness management beyond the office visit, but there is limited evidence for its effectiveness or guidance on how to incorporate it into clinical practice. To test the effectiveness of incorporating PM into clinical practice, we randomly assigned Panel Management Assistants (PMAs) to primary care teams with and without panel management education. METHODS: We conducted an 8-month, cluster-randomized controlled trial at two campuses of the VA New York Harbor Healthcare System. A total of 20 primary care (PC) teams were randomized, which consisted of 51 physicians and 18 nurses serving 8,153 patients with hypertension and/or smoking. Twelve teams were randomized to receive support from a PMA, a college graduate with no clinical training who underwent a one- month orientation that covered basic clinical issues in hypertension and smoking, skill development in panel management, VA's electronic medical record and administrative tools, and motivational interviewing and Brief Action Planning. PMAs systematically reviewed panel data for their assigned teams to identify patients with specific gaps in care. They joined biweekly team meetings to review lists of identified patients and propose PM strategies using a toolkit developed for the study and then conducted patient outreach by phone and mail. Six of the 12 intervention teams also received PM education consisting of five, 20-minute PM education sessions during team meetings. These case-based sessions addressed working in multidisciplinary teams, practicing PM, and leveraging the clinical microsystem. The eight control teams received only monthly data on their smoking and hypertensive patients. Primary outcomes were assessed from the medical record and by patient survey: mean systolic and diastolic blood pressure (BP), proportion of patients with controlled BP, self-reported quit attempts, nicotine replacement therapy (NRT) prescriptions, and referrals to disease management services. PC staff was surveyed before and after the intervention. RESULTS: Regression analysis, controlling for baseline BP and clustering, revealed no significant differences among study arms in mean systolic or diastolic BP values post-intervention. Following the intervention, 90% of smokers reported quitting for at least one day and 64% reported quitting for at least a week. However, there were no significant differences in smoking rates or quit attempts by study group. Patients on intervention teams were more likely to receive NRT for smoking (OR=1.4; 95% CI 1.2-1.6), to enroll in the VA's weight management (OR=1.2; 95% CI 1.1-1.6) or, Telehealth programs (OR=1.7, 95% CI 1.4-2.1), than patients on control teams. Most staff (80%) assigned to an intervention team felt the PMA was a useful resource for their team and 73% wanted to continue working with a PMA. Only 26% said the PMA took too much of their time. Only 40% said they would continue using PM strategies when the PMA left the team. CONCLUSIONS: PM support and education for PC teams improved process, but not outcomes among veterans with hypertension and smoking. This study has important limitations, as it included only veterans, implementation lasted only 8 months with post-intervention data collection interrupted by Hurricane Sandy, and primary care teams were newly formed at study outset. However, incorporating PMAs into PC teams was feasible and highly valued by the clinical staff, and warrants further study.

BEYOND OUR CONTROL: HOW ORGANIZATIONAL CONTEXT IMPACTS PERFORMANCE MEASUREMENT. Molly Harrod1; Jane Forman1; Claire Robinson1; Adam Tremblay2,3; Leo Greenstone2,3; Eve A. Kerr1,3. 1VA Ann Arbor Health Care System, Ann Arbor, MI; 2VA Ann Arbor Health Care System, Ann Arbor, MI;

3University of Michigan, Ann Arbor, MI. (Tracking ID #1937968)

BACKGROUND: The Patient Centered Medical Home (PCMH) model requires a shift from a physician centric to a team-based approach to care, as well as improvement in process measures that reflect PCMH goals. As a result, as primary care (PC) providers are learning to work in teams, they also must learn how to change their clinic processes to meet PCMH-specific performance measures. The Veterans Health Administration (VHA) transitioned over 900 PC sites to a PCMH model beginning in 2010 and has implemented national metrics of PCMH success that include same day access and continuity with the patient's usual provider. One large VHA health system implemented a coaching model to help newly formed PC teams redesign their delivery processes to improve access and continuity measures. This study examines how these PC teams discussed and responded to these measures during their coaching sessions. METHODS: Nine of 20 PC teams were longitudinally observed during coaching sessions (25+ hours) that entailed discussion of both performance measures and processes to improve measure results. Conversations and interactions were recorded via hand-written field notes. We analyzed data using a grounded approach. RESULTS: As PC teams reviewed their access and continuity measures, two predominant, and often

• verlapping, themes emerged: lack of control and lack of consistency across the organization. PC teams felt

they did not have control over processes both within and outside PC. For example, a centralized call center with variable understandings of PC redesign continued to refer patients to the Emergency Department even though PC providers had open appointments within their schedules, thus impacting continuity. A lack of consistency across the organization as a whole was apparent given that many of the other departments did not change their patient care processes to align with PCMH goals. For example, inpatient physicians followed a rule that patients discharged from the inpatient setting needed "follow-up with primary care physician in 7-10 days". Because this 7-10 day mandate meant fewer open slots in a PC provider's clinic, it decreased same day access. Thus,

• rganizational processes were often working against one another resulting measures that were not reflective of

all the changes. CONCLUSIONS: Improving even straightforward performance measures, like same-day access, requires an understanding of the entire practice context. In our study, measurement results that fell short of goals reflected the organizational inconsistency of processes that impacted individual measures. It is important to develop processes designed to meet PACT goals across the organization rather than within each team or department. As more PC practices become part of Accountable Care Organizations and are assessed using these and other performance measures, a better understanding of what is being measured, how these measures reflect patient- centered care, unintended consequences of measures and other organizational actions, and how to align

• rganizational performance goals is needed so that organizational performance reflects true quality of care and

patient-centeredness, and not just performance on individual metrics.