David Keleti, PhD

 To investigate the effect of managed care organization (MCO)-implemented postdischarge engagement, supported by other broadly focused interventions, on 30-day hospital readmissions in 6 at-risk Medicaid populations.

 One-year follow-up analysis of member claims data was performed following an intervention period from January 1, 2014, to December 31, 2014. Postdischarge engagement, supported by additional MCO-initiated interventions, was implemented to reduce 30-day hospital readmissions in Medicaid members having 1 or more dominant chronic conditions. Hospital readmission rates were calculated at baseline and at 1 year post intervention. Bivariable and multivariable generalized estimating equation analysis was used to quantify the likelihood of hospital readmissions.

 Following implementation, postdischarge engagement rates increased significantly, whereas provider follow-up rates remained unchanged. Increased member engagement resulted in statistically significant reductions in weighted readmission rates enterprise-wide (—10.1%; 

 <.01) and in 3 of 6 MCOs (—3.9% to –15.8%; 

 ≤.05) in 2014. Compared with nonparticipants, members who were successfully reached for postdischarge engagement displayed a 33% decrease in 30-day readmissions enterprise-wide (adjusted odds ratio, 0.67; 95% CI, 0.62-0.73) and a comparable decrease (—23% to –39%) in 5 of the 6 MCOs. In this context, greater frequency of postdischarge engagement was associated with proportionally decreased likelihood of readmissions.

 Postdischarge engagement, against the backdrop of multifaceted MCO-implemented interventions, was associated with significantly reduced hospital readmissions in at-risk Medicaid subjects. Reduced likelihood of readmissions was observed at both the enterprise-wide and plan levels in a manner proportional to the frequency of engagement, a novel reported outcome for this population.

Postdischarge engagement, supported by broadly focused stakeholder education and encounters, is effective in reducing hospital readmissions in a frequency-sensitive manner.

At-risk Medicaid beneficiaries who were successfully reached for postdischarge engagement displayed a 33% decrease in 30-day hospital readmissions.

Greater frequency of postdischarge engagement was associated with proportionally decreased likelihood of readmissions, both enterprise-wide and at the individual plan level, a novel reported finding for a Medicaid population.

Health plans can dramatically impact hospital readmission rates through greater postdischarge engagement and a variety of multifaceted interventions aimed at members, providers, and health plan staff.

Hospital readmissions signal gaps in the quality of care provided to patients. Of the 9 million Medicare patient hospitalizations per year,

 almost 1 in 5 are readmitted within a month of discharge and many more return to the emergency department.

 Many such readmissions are caused by inadequate discharge planning, poor care coordination between hospital and community clinicians, and the lack of effective longitudinal community-based care.

Best-practice recommendations to reduce readmissions have largely emerged from analyses conducted on commercial or Medicare fee-for-service populations, whereas relatively few analyses have been published on readmissions in the Medicaid managed care population.

 However, nonpregnant adult Medicaid patients experience readmission rates that are often higher than those experienced by Medicare beneficiaries. Readmission rates for adult (aged 45-64 years) Medicaid patients (22%)

 and those with heart failure (30%) are considerably higher than the corresponding rates in Medicare (16% and 25%, respectively).

 A retrospective analysis of Medicaid beneficiaries living in 19 states reported an average unadjusted 30-day readmission rate of 9.4% (range, 5.5%-11.9%).

 Hospital readmissions represent 12.5% of Medicaid payments for all hospitalizations, averaging $77 million per US state annually.

 Although value-based incentives, financing, and technical assistance can provide powerful drivers to minimize hospital readmissions, most attention in the literature to solving the readmission challenge has focused on Medicare and commercial populations.

However, with increased attention given to Medicaid expansion under the Affordable Care Act, effective approaches to minimize Medicaid hospital readmissions are increasingly being sought. In the pediatric Medicaid population, diseases of the respiratory system were the top cause of readmission, accounting for 21.2% of all readmissions,

 so a preponderance of Medicaid readmissions studies target pediatric asthma. A systematic review of 29 studies identified African American race, public or no insurance, previous admission, and complex chronic comorbidity as risk factors associated with pediatric asthma readmissions.

 The populations analyzed in 14 of the studies included (or were inferred to include) Medicaid beneficiaries.

 Three of the studies reported that the readmission rate of children with asthma insured by Medicaid was higher than that of comparable children with private insurance.

Another systematic review retrieved 21 randomized clinical trials of transitional care interventions targeting chronically ill adults and further identified 9 interventions that demonstrated positive effects on hospital readmissions—related measures.

 Many of the successful interventions shared similar features, such as assigning a nurse as the clinical manager or leader of care and including in-person home visits to discharged patients. Five of the studies had interventions that could be described as discharge management plus follow-up,

 led to reductions in readmissions through at least 30 days after discharge. The majority of these interventions were performed on elderly populations, and only 1 study consisted of a sizable Medicaid cohort.

In 2014, AmeriHealth Caritas Family of Companies, a Medicaid managed care organization (MCO), implemented multifaceted enterprise-wide initiatives—including enhanced postdischarge engagement&mdash;to reduce 30-day all-cause readmission rates for plan members having 1 or more dominant chronic conditions, or a single condition of moderate chronic asthma, during the 2013 baseline period. We evaluate the impact of these initiatives in reducing hospital readmissions over a 1-year period enterprise-wide and across 6 MCO affiliates in 4 states (Pennsylvania, Louisiana, South Carolina, and Nebraska) and the District of Columbia (DC).

This was a 1-year follow-up prospective study. The study cohort included Medicaid members aged 5 to 64 years who had 1 or more dominant chronic conditions, or a single condition of moderate chronic asthma (selected for high prevalence and impactability in the Medicaid population), from January 1, 2013, to December 31, 2013. Medicaid—Medicare dual-eligible members (≥65 years) were not served during this time period. Subjects were members of 6 Medicaid MCOs serving 4 states (southeastern Pennsylvania [PA01], Lehigh/Capital–New West Pennsylvania [PA02], Louisiana [LA], South Carolina [SC], and Nebraska [NE]) and DC. Baseline index inpatient admissions were included from January 1, 2013, through December 31, 2013, with claims paid through March 31, 2014, or inpatient admission incurred before January 1, 2013, or discharged after December 1, 2012, with claims paid through March 31, 2014. Baseline 30-day all-cause inpatient readmissions were included if they were incurred between January 1, 2013, and December 31, 2013, with claims paid through March 31, 2014, and the admission date was less than 30 days from prior index inpatient discharge. For measurement year 2014, the same logic was followed, with the admit date or discharge date advanced 12 months forward.

Hospital Readmission Identification Logic

Hospital readmissions were extracted from claims data housed in AmeriHealth Caritas’ data management platform. Index admission was defined as an acute admission or maternity delivery. The outcome variable of 30-day all-cause readmission was based on acute admissions (but excluding maternity deliveries) incurred within 30 days after discharge of prior index admissions.

A variety of initiatives targeting members, providers, and MCO associates were implemented to reduce 30-day all-cause hospital readmissions and expenditures in at-risk Medicaid members (

), ranging from nonclinical interventions (eg, telephonic care management) to stakeholder communications (eg, member mailings, continuing medical education webinars for providers). Most interventions were general in nature, but several were member-specific and tiered to the member’s risk level (eg, telephonic vs in-person contact during postdischarge planning). Pharmacy-based initiatives were implemented to improve medication adherence in partnership with the pharmacy benefit manager PerformRx.

Members, especially those with the highest risk of readmission, were contacted following discharge, as were hospitals with the highest readmission rates to discuss ways to reduce member readmissions. Care management and care coordination decisions were made based on each member’s diagnosis, readmission risk factors, and available home supports. Individual health management barriers were identified by speaking with the patient and his or her family and provider. Interventions focused on ensuring that the member understood discharge instructions (including medication schedules) and on facilitating postdischarge provider visits.

Thirty-day all-cause readmission rates are expressed as utilization per 1000 members per year (PKPY) in each MCO.

The percentage change of readmission rate per MCO is the quotient value of the difference in readmission rates between 2014 and 2013 divided by the readmission rate in 2013. The overall readmission rate (in PKPY) is the quotient value of the above formula summed for all 6 MCOs. The final rate is a weighted average of readmission rates across all 6 MCOs at baseline and the study period; plans having more participants contributed a proportionally greater rate to the overall average versus plans with fewer participants, better reflecting overall performance.

Also reported was the number of successful member calls within 30 days of the discharge date for the index admission (from January 1, 2014, to November 30, 2014) or within the days between the index discharge date and readmit date, whichever was shorter. A successful member call was only logged when an associate spoke directly with the member or his/her parent or guardian during the outreach, but not if the associate left a message or spoke to someone else at the residence.

 test was performed to compare differences in pre- versus postintervention engagement rates and postdischarge follow-up rates with providers within 7, 14, and 30 days of inpatient discharge in 2013 and 2014. The Poisson distribution with GENMOD procedure (PROC GENMOD) was used to test readmission rate differences pre- and post intervention (details in 

The number of successful member calls was grouped as a categorical variable with 4 groups (0, 1, 2, ≥3 calls) and as a dichotomous variable (0 vs ≥1 calls). Chi-square tests and Cochran-Armitage trend tests were used to compare the association between the engagement (≥1 call vs no calls) and readmissions and between the number of successful calls (0, 1, 2, ≥3 calls) and readmissions.

To quantify the likelihood of readmission among members with a different number of successful calls in 2014, the index discharge became the study unit. Because some members encountered multiple discharges in the study period, the generalized estimating equations (GEE) model was employed to account for repeated measures for 1 subject. The model used a binomial likelihood function, a logit link function, and an independent correlation matrix. Empirical standard error estimates were used to estimate the standard errors. Odds ratios (ORs) and 95% CIs were then calculated as the exponentials of the parameters (OR = 

). Two types of models with different independent variables were developed: 1 dichotomous (≥1 calls vs no call) and the other categorical (1, 2, ≥3 calls vs no call). For both models, bivariable GEE analysis was first performed to explore the unadjusted effect of successful calls on preventing readmission. Multivariable GEE analysis was then used to quantify the independent effect after controlling for sociodemographic variables like age, gender, race, aid category, and risk scores.

All analyses were performed using SAS EG 7.1 (SAS Institute; Cary, North Carolina). A significance level of 

 <.05 was deemed statistically significant for all comparisons.

Medicaid subjects (N = 149,748) in 6 MCOs had 1 or more dominant chronic conditions (range, 54.41%-75.39%) or a single condition of moderate chronic asthma (range, 24.61%-45.59%) in 2013 (

). Clinical Classifications Software (CCS Grouper; Rockville, Maryland) clustered the top 10 diagnosis categories of the study group at baseline, accounting for more than 45% of all hospital readmissions enterprise-wide, which were (in descending order) sickle cell disease, diabetes with complication, congestive heart failure, complicated procedures, asthma, chronic obstructive pulmonary disease, septicemia, pneumonia, chest pain, and skin infection. The majority of subjects in all MCOs were female (range, 51.97%-58.42%), and a plurality were younger than 18 years (range, 27.93%-58.06%; primarily because of asthma). In other respects, the MCOs comprised a diverse population.

Postdischarge Engagement and Provider Follow-Up 

The rates of MCO-initiated postdischarge engagement (

) were compared in 2013 and 2014. Postdischarge engagement rates within 7, 14, and 30 days after discharge increased significantly enterprise-wide from 2013 to 2014 (15.07 percentage points [from 10.18% to 25.25%], 18.90 percentage points [from 14.18% to 33.08%], and 21.37 percentage points [from 17.74% to 39.11%], respectively; 

<.001 for all) and for all MCOs except 1 (

 = .018). Conversely, provider follow-up rates remained statistically unchanged.

The hospital readmission rates following discharge were also compared in 2013 and 2014 (

). The enterprise-wide weighted 30-day hospital readmission rate significantly decreased from 41.30 PKPY to 37.11 PKPY (—10.1%; 

 <.01) between 2013 and 2014. Postdischarge hospital readmission rates decreased in every MCO (range, —3.9% to –32.2%), but the reduction was only statistically significant in 3 of the 6 MCOs (PA01, 

 <.01). The enterprise-wide weighted hospital readmission rate for members with asthma alone significantly decreased as well (—20.7%; 

Postdischarge Engagement Prevents Hospital Readmissions 

To investigate the impact of successful postdischarge engagement on readmissions, hospital readmission rates of members who were successfully engaged post discharge at least once (≥1 successful call) were compared with unengaged members (no successful calls; 

). Members who were successfully engaged post discharge had significantly lower readmission rates enterprise-wide than unengaged members (14.5% vs 19.1%; 

 <.001). Even following adjustment with confounding variables of age, gender, race, aid category, and risk scores, postdischarge engagement was correlated with a 33% decrease in hospital readmissions (adjusted OR, 0.67; 95% CI, 0.62-0.73). A comparable decrease was observed in every MCO (range, —23% to –39%; adjusted OR, 0.61-0.77), except for NE, the plan with the smallest sample size.

We next examined whether an increasing number of successful postdischarge calls was associated with a proportionately decreasing likelihood of hospital readmission (

). Enterprise-wide, members successfully engaged post discharge with 1, 2, and 3 or more contacts all had significantly lower readmission rates than unengaged members (16.3%, 15.4%, and 11.3% vs 19.1%, respectively; 

 <.001 for all). Even following adjustment with confounding variables of age, gender, race, aid category, and risk scores, increasing the number of postdischarge engagements from 1 to 2 to 3 or more successful contacts displayed decreasing likelihood of hospital readmissions (adjusted ORs, 0.80, 0.74, and 0.54, respectively). This observation was consistently detected in every MCO except for NE; however, the CI for 1 versus 0 and 2 versus 0 successful calls overlapped with the ordinate at 1.0 (indicating no change in OR) for 3 and 2 MCOs, respectively. In other words, 1 postdischarge engagement was insufficient to drive a decreased likelihood of readmission rates in 3 of the 6 MCOs (LA, SC, and DC), and even 2 postdischarge engagements were insufficient in 2 MCOs (LA and SC). By contrast, 3 postdischarge engagements were sufficient to decrease the likelihood of readmission rates in all 5 MCOs for which the OR could be calculated.

Ineffective transitions of care following hospital discharge are a major challenge in population health, where limited windows of opportunity exist to engage patients in follow-up care.

 Postdischarge engagement, supported by other broadly focused MCO-led nonclinical interventions, was examined for its impact on 30-day hospital readmissions in at-risk Medicaid subjects. Following implementation, postdischarge engagement rates increased significantly enterprise-wide (while provider follow-up rates remained unchanged), resulting in a statistically significant reduction in weighted readmission rates enterprise-wide, accompanied by a 4.8% reduction in associated expenditures. Members successfully engaged post discharge displayed decreased likelihood of readmission compared with those unengaged in 2014. Moreover, greater frequency of postdischarge engagement was associated with proportionally decreased likelihood of readmissions, a novel reported outcome for this population.

Although all 6 MCOs reported declines in readmission rates, 3 MCOs reported nonsignificant declines (

 >.05). Two of these 3 MCOs had the lowest baseline readmission rates, while eliciting further reductions in postdischarge readmission rates with the highest magnitude declines among all MCOs (NE, −32.2%; DC, −19.9%). Therefore, the nonsignificance was largely due to their small population sizes relative to other MCOs (NE and DC had 1% and 5%, respectively, of the total study population). Moreover, the third “nonsignificant” MCO (PA02) showed a borderline 

 value of .09. Finally, inter-MCO differences in member sociodemographic profiles (racial composition, rurality, chronic disease-to-asthma ratio, and non-SSI status) and degree of implementation of the assorted postdischarge engagement processes (especially those dependent on the local health ecosystem) likely contributed to inter-MCO variations.

Reasons for this study’s successful outcomes are varied. First, postdischarge engagement by care managers established a link between member and MCO, providing a touchpoint for addressing postdischarge barriers. Postdischarge outreach addresses issues relating to medication possession and equipment/home care services, comprehension of discharge instructions, and reporting of signs and symptoms. The proportionally lower likelihood of hospital readmission with increasing telephonic engagements in some cohorts was striking. Care managers representing MCOs—associated by members with the quality of medical care they receive&mdash;frequently hear health-related complaints. At minimum, telephonic engagement fosters member awareness of their health-related needs. Care managers are trained to spontaneously identify potentially serious issues and assist in addressing them before they require emergency care, an impact greatly amplified when: (1) contacting members with chronic and comorbid conditions (more likely to be enrolled in care management) and (2) completing validated health-related assessments on members to uncover additional assessment-focused concerns (Glenn Hamilton, MD, personal communication, December 8, 2017). Finally, members develop a stronger relationship with care managers who contact them more regularly.

Lastly, nonclinical initiatives helped establish a framework for implementing and communicating postdischarge solutions among members, providers, and MCO associates. Many successful initiatives were later adopted into the MCO’s system of care. Overall provider engagement remained steady during most follow-up periods in most plans. Providers often lack dedicated staff for effective postdischarge engagement and may be unaware that a patient had a hospital admission. Conversely, MCOs primarily become aware of their members’ hospital admissions upon processing their medical claims to reimburse hospitals for services rendered, so improved MCO—provider communication regarding patient status alerts both parties of unanticipated acute hospitalizations, follow-up appointment status, and existing care gaps. MCOs can supply de facto “back-office support” in patient engagement to provider offices with limited resources. Successful MCO-led member engagement is an ongoing challenge across all health plan programs and requires creative solutions, like initial outreach via SMS texting. Thus, additional investigations regarding the effectiveness of postdischarge engagement and other nonclinical interventions are warranted.

Articles

Postdischarge engagement of at-risk Medicaid beneficiaries in 6 health plans resulted in significant reductions in hospital readmissions at rates proportional to the frequency of engagement.

Postdischarge Engagement Decreased Hospital Readmissions in Medicaid Populations

To evaluate the effects of a collaborative pharmacy benefits manager (PBM)/ health plan—administered drug therapy management (DTM) program on healthcare utilization and costs in patients with diabetes treated with polypharmacy.

 Retrospective quasi-experimental design with comparison group.

This DTM program was a collaborative effort between the PBM, PerformRx, and the care management departments of Keystone First (KF) and AmeriHealth Caritas Pennsylvania (ACP) care management departments, targeting patients with diabetes using 

15 medications. Pharmacists reviewed member profiles and made evidencebased prescriber and patient interventions, working directly with prescribers and indirectly with members, via care managers. Care managers provided additional services not otherwise within the scope of DTM. The study group consisted of 954 DTM participants reviewed by a pharmacist between November 1, 2010, and July 31, 2011. The control group consisted of 810 matched DTM participants not reviewed by a pharmacist.

 Intervention acceptance rates for KF and ACP were 33% and 26%, respectively. The study group demonstrated lower inpatient admissions and emergency department utilization rates, although only the KF study group inpatient admission rate achieved statistical significance (76.4%; 

 = .0002). The study groups realized statistically significant total cost savings (pharmacy + medical) compared with their corresponding control groups (47.8% KF, 

 = .0497) despite non-statistically significant increases in pharmacy costs.

 A collaborative pharmacist-driven DTM program with a care manager—executed patient outreach component results in reduced hospital utilization and significant healthcare cost savings.

A collaborative drug therapy management (DTM) program with modest acceptance rates effectively reduces inpatient admissions and generates significant cost savings.

Overall acceptance rates for pharmacist-recommended DTM interventions are likely to range from 26% to 33%.

Hospital utilization can be substantially reduced in a successful DTM program.

Pharmacy-related costs would be expected to increase with the implementation of a successful DTM program but are outweighed by medical cost savings.

According to the American Pharmacists Association, drug therapy management (DTM) is a service that optimizes therapeutic outcomes by way of pharmaceutical interventions intended to (1) elicit changes in drug regimens; (2) reduce the incidence of adverse drug events; and (3) improve adherence. Formerly an underutilized component of the healthcare team approach to treating patients, implementation of DTM services has been facilitated by a payment framework created by the Medicare Prescription Drug, Improvement, and Modernization Act of 2003.

DTM is frequently a stand-alone service consistent with a traditional, siloed approach to healthcare. Although an integrated approach to healthcare is preferred, DTM as a component of a segmented healthcare system has demonstrated impressive quality improvements and cost savings. The Minnesota Medicaid Medication Therapy Management program realized healthcare cost savings of $3768 per patient per year between 2006 and 2007.

 The North Carolina ChecKmeds medication therapy management program for more than 15,000 Part D seniors saved an estimated $13.2 million in healthcare costs in 2010.

 The Iowa Medicaid Pharmaceutical Case Management program demonstrated a significantly improved mean Medication Appropriateness Index score among participants compared with baseline (P <.001) over a 9-month period between 2000 and 2001.

 Managed care organizations (MCOs), occasionally employing pharmacy benefits managers (PBMs),4 have also increasingly incorporated DTM services into benefit offerings, demonstrating positive outcomes.5,6 While DTM service providers are primarily pharmacists, a pharmacist is not a prerequisite for a DTM program. In fact, in 10% of 2012 Medicare Part D programs, a case manager provided DTM-type services.

Not all patients, however, benefit from DTM services, and intervention should focus on those at highest risk. Comorbidities such as cardiovascular disease, hypertension, and nephropathy

 frequently accompany a diagnosis of diabetes, and patients with multiple comorbidities are often managed with polypharmacy. Polypharmacy increases the risk of medication-related problems (MRPs), reducing population health and driving up health care costs, a problem complicated by the prescription of more medications than clinically warranted. Polypharmacy also promotes suboptimal adherence,

 resulting in underutilization of preventative, lower-cost primary care services and overutilization of higher-cost emergency department (ED) services.

 Thus, despite lower medication-related costs from low adherence and reduced pharmacy utilization, costs are transferred to the medical sphere via less effective emergency care, thereby increasing overall healthcare costs.

DTM is a valuable service proven to improve healthcare quality and reduce costs. Various DTM delivery models exist. A literature search revealed a lack of DTM programs that incorporate a care manager in addition to a pharmacist (see 

). The value of DTM services may be enhanced with an integrated approach that includes a care manager. This study was designed to evaluate the effects of a PBM/health plan—administered, collaborative DTM program that incorporates both a pharmacist and a care manager targeted to high-risk patients with diabetes treated with polypharmacy. We were primarily interested in determining whether the DTM program would reduce medical utilization and total costs (pharmacy + medical). We hypothesized that DTM intervention would reduce medical utilization, as measured by ED visits and inpatient admissions, and generate total cost savings, despite a predicted increase in pharmacy costs.

PerformRx, the fully owned PBM of AmeriHealth Caritas Family of Companies (ACFC), serves just under 3 million health plan members. Partnering Medicaid MCOs and fully owned ACFC subsidiaries Keystone First (KF) and Ameri- Health Caritas Pennsylvania (ACP) serve about 290,000 and just over 126,000 members, respectively, as of January 2014. These 3 entities jointly developed a collaborative, Utilization Review Accreditation Commission—accredited DTM program to improve member care. PerformRx was responsible for the pharmacy-based management component of the DTM program. KF and ACP supplied the DTM program participants and executed the care management (CM) component.

This retrospective study was designed as a quasi-experiment with comparison group. The targeted participants were patients with diabetes (

International Classification of Diseases, Ninth Revision

, codes for diabetes [250.xx] or 1 diabetes medication within the previous year) being treated with polypharmacy (unique generic code number count of 

15). This patient population was selected because the likelihood of comorbidities with diabetes and polypharmacotherapy put these patients at high risk for MRPs and nonadherence.

The study population analyzed consisted of 954 DTM participants (n = 690 KF, n = 264 ACP) reviewed by a pharmacist for DTM services. The matched control group consisted of 810 DTM participants (n = 600 KF, n = 210 ACP) not reviewed by a pharmacist for DTM services, but who may have received autonomous CM services. The baseline period (November 1, 2009-October 31, 2010) was followed by a 1-year measurement period (November 1, 2010-October 31, 2011). Members enrolled in the plan for less than 6 months during baseline or measurement periods were excluded. Additionally, members reviewed by a pharmacist in the final 3 months of the measurement period (August 1, 2011-October 31, 2011) were excluded because recommendations made during that time would not have had the standard 3-month time frame allowed for follow-up and outcome determination. Analysis was performed 6 months postmeasurement, allowing sufficient lag time to receive medical claims.

Relative risk scores quantify financial implications of morbidity based on prior healthcare utilization.

 Prospective risk scores were assigned to participants based on previous claims, gender, and age, and used to randomly select a comparison group with a comparable risk score. Since risk scores are continuous variables, 1:1 matching of study-to-control group risk scores was impossible. Therefore, we applied stratified random sampling of risk scoring from the 10th percentiles of the study group as strata. Cutoff points for these 10th percentiles were applied to the control groups to create 10 strata. An equal number of cases from each of the 10 strata was randomly selected, generating a comparable distribution of risk scores in control and study groups. The comparability of these groups was validated with an independent, 2-sample t test.

). DTM pharmacists reviewed adherence, gaps in care (GICs), and First DataBank (FDB) clinical decision-support modules, and prepared prescriber and member interventions. Adherence was assessed for both short-term (eg, member overdue refilling medication) and long-term (eg, member frequently misses doses over an extended time frame) issues. Interventions to improve adherence included recommendations to simplify regimens and reduce side effects and drug interactions.

 GICs are proprietary algorithms that identify when members have not met standard-of-care guidelines (eg, diabetic not self-monitoring blood glucose [SMBG]). FDB modules identify safety concerns such as drug interactions, duplicate therapies, and improper dosing.

Pharmacists provided evidence-based recommendations directly to prescribers and indirectly to members via care managers, referring to a list of medications considered "urgent," jointly compiled by PerformRx and KF/ACP. Urgent interventions directed to prescribers received a phone call in addition to the standard letter/fax, and those directed to members received expedited care manger outreach within 1 day compared with the standard 20 days. KF and ACP employ the same style of CM—a blended model of CM and disease management managed by a single care manager. Care managers provided additional services not otherwise within the scope of DTM, such as health coaching, education, transportation assistance, and prescription refill/transfer/authorization assistance. Members actively engaged in CM were counseled by their respective care managers, thereby leveraging existing care manager—member relationships and streamlining contact efforts. Members not actively engaged in CM were counseled by an urgent response care manager, thereby expanding available resources to include CM services. DTM pharmacists were available to assist care managers with counseling.

After 20 days from the date a member intervention was identified, and 90 days from the date a prescriber intervention was identified, the pharmacist and technician, respectively, followed up to determine the outcome of the outreach. Pharmacists reviewed care manager outreach notes and technicians reviewed claims data to determine outcomes. Outcomes were flagged as Accepted, Not Accepted, Rejected, Unknown, or Not Applicable. In the event a recommendation to add or change therapy was accepted, the pharmacist conducted an additional review for medication appropriateness and adherence and tasked a member outreach to the care manager to assess efficacy and side effects.

Dependent variables examined to demonstrate the effectiveness of the DTM program included number of ED visits and inpatient admissions per 1000 members, as well as per member per month (PMPM) pharmacy, medical, and total costs. These variables were populated with the health plans’ claims data. The independent variable analyzed was percent participation in the DTM program.

Because variations in risk can be important cost and utilization confounders, Diagnostic Cost Group (DxCG, Boston, Massachusetts) models were applied using risk adjustment methodology to establish an empirically valid measure of expected resource use. Mean risk scores for KF study and matched control groups were 695 and 702, respectively. Mean risk scores for ACP study and matched control groups were 660 and 643, respectively (

For each member, the dependent variables during baseline and measurement periods were summed separately and divided by the number of member months in the corresponding time period. Healthcare utilization and costs for study and control populations comparing baseline and measurement time periods were evaluated using the nonparametric Wilcoxon signed-rank test, selected because the data were matched pairs and the variables were not normally distributed.

Difference-in-differences (DID) analyses were used to compare program effects on utilization and costs study and control groups. Statistical significance was set at α = .05. Data manipulation and analysis were performed using SAS EG 5.1 (SAS Institute, Cary, North Carolina).

The percentages of KF and ACP study group members that received pharmacist intervention were 92.3% and 95%, respectively. Pharmacists reviewed without intervening on a subset of study group subjects who had no MRPs. The control group did not receive pharmacist review. Study, control, and total plan population demographic data are displayed in Table 1. The total number of diabetic members in KF and ACP were 11,372 and 3655, respectively.

The disease burden of the study and control groups was approximately 7-fold greater than the health plan average, as measured by differences in DxCG risk scores. In both plans, females and older members (aged 45-64 and aged 65+ groups) were over-represented in the study and control groups compared with the total plan population. A more aged DTM participant population is consistent with the greater likelihood of polypharmacy at more advanced ages. Racial and ethnic demographics between the 2 plans were distinct and consistent with regional demographics. African Americans were underrepresented and multiracial members were over-represented in the study group compared with the control and total plan population in ACP. In KF, mean ages were disparate between study and control groups (53 vs 67 years, respectively), while the median ages were similar (54 vs 55 years, respectively; Table 1).

); the top 3 intervention types and their proportion of all recommendations are: maintenance medication nonadherence (eg, statins): 37% to 41%; suboptimal therapy (eg, GICs): 21% to 22%; lack of medication monitoring/selfmonitoring (eg, SMBG): 14% to 16%.

Acceptance rates for diabetes-specific interventions for KF and ACP were 33% and 26%, respectively. SMBG and medication adherence recommendations had higher acceptance rates (38%-54% KF, 29%-57% ACP) than recommendations for drug-drug interactions (33% KF, 18% ACP) or addition of medications (12%-13% KF, 10%-11% ACP).

To demonstrate the effectiveness of the DTM program in limiting MRPs for study group participants, hospital utilization— reflected in the number of ED visits and inpatient admissions per 1000 members&mdash;was assessed and reported in 

, respectively. The differences between measurement period visits/admissions and baseline period visits/ admissions were compared.

In both plans, a statistically significant increase in ED visits was observed in the control group (16.9% KF, 

 = .0107); however, the changes observed in the study group were not statistically significant (—2.6% KF, 

 = .4883; Figure 1). DID analysis revealed that the reductions in ED visits at 19.5% (n = 343; 

 = .5728) for ACP were not statistically significant (

A statistically significant increase in KF inpatient admissions was observed in the control group (66.0%; 

 <.0001), while a statistically significant decrease was observed in the study group (—10.4%, 

 = .0008; Figure 2). An increase in ACP inpatient admissions was observed in both the control and study groups (43.8% and 7.9%, respectively), although neither value achieved statistical significance (

 = .4196, respectively). DID analysis revealed that the reduction in inpatient admissions for KF was statistically significant at 76.4% (n = 856; 

 = .0002), but not statistically significant for ACP at 35.9% (n = 222; 

To determine the effectiveness of the DTM program in controlling study group costs, PMPM pharmacy, medical, and total costs were compared (

). Pharmacy costs increased significantly in the control group (19.9% KF, 25.0% ACP; both

 < .0001) and in the study group (10.1% KF, 

 = .0155). From 2010 to 2011, average pharmacy costs PMPM for KF and ACP increased by 14.1% and 12.4%, respectively. However, DID analysis revealed that pharmacy cost increases were not statistically significant (9.8% KF, 

Medical costs increased significantly in the control group (67.4% KF, 

 = .0016), but not the study group (1.6% KF, 

 = .1797). From 2010 to 2011, average medical costs PMPM for KF and ACP increased by 4.6% and 0.7%, respectively. When comparing the differentials between study and control groups, the reduction in medical costs for KF was statistically significant (65.8%; 

 = .0031), but not statistically significant for ACP (73.8%; 

Total costs increased significantly in the control group (52.0% KF, 

 <.0001), but not the study group (4.2% KF, 

 = .3785). From 2010 to 2011, average total costs PMPM for KF and ACP increased by 6.6% and 9.4%, respectively. According to the DID analysis, the reductions in total costs for KF (47.8%; 

 = .0497) were statistically significant (Table 2).

Managing healthcare costs while maintaining a high level of patient care is a national priority. An estimated $289 billion is spent annually on drug-related morbidity (13% of total healthcare expenditures), primarily resulting from poor medication adherence, suboptimal prescribing, incorrect drug administration, and inaccurate diagnosis.13 The use of pharmacists and care managers as part of a health plan-administered DTM program can help to prevent and correct MRPs, reduce medical utilization, and decrease total costs. We successfully implemented a pilot DTM program in members with diabetes treated with polypharmacy resulting in reduced hospital utilization at substantial cost savings for 2 health plans. The program was comparably effective in 2 disparate populations with high illness burdens. African Americans were over-represented in the urban KF population, while Hispanics and multiracial members were over-represented in the rural ACP population, compared with the national average.

Adherence and self-monitoring recommendations accounted for over 50% of all DTM interventions while also demonstrating the highest acceptance rates. We speculate that higher acceptance rates were achieved because these intervention types involved both prescriber-directed pharmacist outreach and patient-directed care manager outreach. Although not statistically confirmed, the duality of this approach likely increased the recommendation acceptance rate. By contrast, recommendations that were solely prescriberdirected, such as additional therapy recommendations, demonstrated lower acceptance rates.

Despite statistically significant total cost savings in both health plans, only inpatient utilization in the KF study group was reduced in a statistically significant manner. A number of reasons may account for this discrepancy, particularly the selection criteria. In line with ACFC’s goal to improve member health, members with diabetes were reviewed in order of degree of healthcare utilization. The study group, therefore, contained more high utilizers compared with the control group. Matched control subjects were later selected based upon comparative risk assessments, which have their own inherent limitations. Therefore, based on the statistically significant total cost savings in both plans, we surmise that the actual utilization benefits of the DTM program were greater than those reflected by the data.

Interestingly, pharmacy cost escalation was observed in the control group as well as in the study populations. We suspect that most clinicians were unaware whether their patients were nonadherent or poorly adherent, so when the patient’s diabetes was not controlled on maintenance medication (ie, metformin), the physician switched to, or added, a more expensive medication (eg, TZD), thereby increasing the cost for the control groups.

Several additional limitations are apparent in this study. First, the impact of intervention on patients who were fully or partially engaged in CM versus those not engaged in CM was not determined. Members not fully engaged in CM may be less receptive to DTM intervention and may have demonstrated lower acceptance rates in a subgroup analysis. In the KF cohort, 43% of the study group was actively care managed at some point during the measurement period compared with only 35% of the control group. In the ACP cohort, the corresponding values were 31% and 25%, respectively. Additionally, the impact of the DTM program on quality outcomes, such as Healthcare Effectiveness Data and Information Set (HEDIS) measures, was not determined (limitations that will be addressed in future studies). Finally, because total costs were calculated by summing only ED visits, inpatient admissions, primary care provider visits, and specialist visits, they are not a comprehensive total of medical costs. This calculation excludes the cost of medications associated with a procedure, laboratory and durable medical equipment claims data, out-of-pocket costs, and costs to the behavioral health MCO (if present; ACFC is not privy to this information, so is unable to account for it in statistical analysis).

The ACFC DTM program is a collaborative initiative integrating DTM services into existing CM programs. This pilot study demonstrated positive outcomes in both utilization rates and healthcare costs. First, the KF study group demonstrated a statistically significant reduction in inpatient admissions compared to the control group; the reduction observed in ACP participants was not statistically significant. Comparative outcomes for ED utilization were less dramatic and not statistically significant for either plan. Second, optimizing medication regimens for study group participants generated statistically significant reductions in total healthcare costs compared with the control group in both plans. Thus, a collaborative DTM program can be a successful component of managed care benefit offerings and contribute to achieving the goal of providing better care and superior quality at a lower cost.

PerformRx, Philadelphia, PA 19113 (LB, AW, MS, MT); The AmeriHealth Caritas Family of Companies, Philadelphia, PA 19113 (EJB, DK, KEM, SF, CJ, STT, ADG).

Work on this manuscript was supported by PerformRx and the AmeriHealth Caritas Family of Companies.

Drug therapy management implementation in 2 health plans resulted in significant cost savings and modest to significant reductions in emergency department visits and inpatient admissions among patients with diabetes.

David Keleti, PhD

Articles

Postdischarge Engagement Decreased Hospital Readmissions in Medicaid Populations

Collaborative DTM Reduces Hospitalization and Healthcare Costs in Patients With Diabetes Treated With Polypharmacy