Richard L. Sheer, BA

To evaluate the impact of a collaborative effort of a Medicare Advantage and prescription drug (MAPD) plan and community pharmacies to improve vaccination rates for pneumonia and influenza.

 This quasiexperimental, cluster-randomized intervention study used MAPD data to assess the impact of community pharmacists on vaccination rates. Pharmacies in specific regions (districts) were randomly assigned to intervention or control groups. Intervention pharmacies received reports of patients with a gap in influenza (aged 19-89 years) and/or pneumococcal (aged 65-89 years) vaccinations based on medical and pharmacy claims history. Vaccine-naïve patients were offered vaccinations.

 The vaccination rates for the previously vaccine-naïve patients utilizing intervention and control pharmacies were compared 6 months post randomization. Inverse probability weighted hierarchical generalized linear models determined the odds of receiving pneumonia and influenza vaccinations for intervention and control groups, controlling for baseline clinical and demographic characteristics.

 Intervention and control groups had similar ages in the pneumococcal older-adult cohort (mean age, 73.0 vs 73.4 years, respectively; 

= .1255). The intervention group was older than the control group in the influenza cohort (mean age, 67.7 vs 66.4 years, respectively; 

= .0006). Slightly more than half of each cohort were women, and the proportion of women was not significantly different between the intervention and control groups in each cohort. In multivariable analyses, intervention pharmacies were associated with higher odds of delivering pneumococcal (odds ratio [OR], 1.91; 95% CI, 1.26-2.87) and influenza (OR, 2.18; 95% CI, 1.37-3.46) vaccinations than control pharmacies.

 A health plan–enabled, pharmacist-led intervention was effective in increasing the number of older adults receiving pneumococcal vaccination and individuals receiving influenza vaccination.

Patients whose pharmacies receive notification of their immunization gap have twice the odds of receiving immunizations compared with those whose pharmacies do not receive the notification.

A Medicare Advantage health plan provided community pharmacists with selective lists of patients who have an immunization gap for pneumococcal or influenza vaccinations, followed by confirmation from patients and an invitation to receive the needed vaccination at the pharmacy.

The intervention had a significant impact on the rate of pneumococcal and influenza vaccinations. During the 6-month postintervention time period, the odds of receiving either a pneumococcal vaccination or an influenza vaccination for the intervention group were more than double that of the control group (both 

Vaccine-preventable diseases such as influenza and pneumococcal infections can lead to significant morbidity and mortality. Influenza vaccines are effective in preventing hospitalization and death in individuals with high-risk medical conditions.

 Similarly, pneumococcal vaccines can prevent approximately 16,000 deaths annually and can save $1.8 billion in health care costs related to hospitalizations, outpatient visits, and medications used to treat older adults (≥ 65 years) with pneumonia.

 Nevertheless, adult vaccination rates remain low for most routinely recommended vaccines, and they are well below Healthy People 2020 targets.

Myriad strategies have been described to close gaps in vaccination rates. A systematic review found that increasing community demand (eg, patient education, educational letters/flyers), enhancing vaccination access (eg, home visits, low-cost or free vaccines), and provider- or system-based interventions (eg, financial incentives, reminders, educational outreach) were associated with improvements in influenza vaccination rates.

 Among examples of provider-based interventions, pharmacist-driven interventions have improved immunization rates. For example, a statewide, pharmacist-driven pneumococcal vaccination educational outreach program led to higher vaccination rates and lower pneumococcal infection rates during the study period.

 Another pharmacist-managed vaccination campaign in high-risk patients at a rural primary care clinic substantially increased influenza vaccination rates among patients 65 years and older, from 44.6% to 70.9% (

 A peer-to-peer outreach intervention consisting of an educational mailing followed by tracking and outreach by senior volunteers was effective in increasing influenza and pneumococcal vaccination rates.

 Three additional systematic reviews found that pharmacists played significant roles in screening, educating, and immunizing, thereby improving pneumococcal and influenza vaccination rates.

Although there have been improvements in vaccination rates over time, only 59% of older adults have received at least 1 of the pneumococcal vaccinations, as seen in the pneumococcal vaccination rates of Medicare beneficiaries published by the CDC (2009-2017).

 Furthermore, a significant gap still exists regarding influenza vaccination within this population, with approximately 45% of Medicare beneficiaries receiving the influenza vaccine during the 2017-2018 season.

 In an attempt to continue to improve the pneumococcal and influenza vaccination rates among individuals enrolled in Medicare Advantage and prescription drug (MAPD) plans, this pilot study designed, implemented, and evaluated a program that identified and communicated immunization gaps to community pharmacists in a pharmacist-led intervention regarding these vaccinations.

This quasiexperimental, cluster-randomized intervention study examined the impact of providing community pharmacists with information related to patient history of pneumococcal and influenza vaccinations based on both medical and prescription claims. Pharmacies in specific regional areas (districts) consisting of approximately 25 to 30 pharmacies with comparable patient characteristics were randomly divided into 2 groups, with a balanced distribution of urban/rural, age, and sex. The intervention group consisted of 1 urban and 1 rural district in the Lexington/Frankfort, Kentucky, area, and the control group consisted of 1 urban and 1 rural district in the Louisville, Kentucky, area. The study was reviewed and approved by the Chesapeake and Advarra (formerly Schulman) institutional review boards.

Data were provided by Humana Pharmacy Solutions for individuals enrolled in MAPD plans with Humana Inc and who filled prescriptions at the participating pharmacies. Pneumococcal and influenza quality measures were configured into a cloud-based performance information management platform (Electronic Quality Improvement Platform for Plans & Pharmacies [EQuIPP]), tested, and implemented to report rates of immunizations. Patients were attributed to specific pharmacies based on where they filled a majority of their prescriptions. In the event of a “tie,” the patient was assigned to the pharmacy where they filled a prescription most recently.

Pharmacies designated in the intervention group received, via the EQuIPP application, active reporting of individuals who had a gap in influenza and/or pneumococcal vaccinations from their medical and pharmacy claims history. Active reporting included both monthly performance tracking for the immunization measures and patient-level details regarding those who had not received the applicable immunization during the measurement period. Patients were queued to receive a consultation from the pharmacist to determine if a vaccination was appropriate when they came into the pharmacy to fill a prescription (ie, the pharmacist asked whether they had received the vaccination that was not documented in claims, and if not, shared the indications as stated in the guidelines and confirmed there were no contraindications). Training for pharmacists was conducted via on-demand teleconference beginning December 23, 2016. Control group pharmacies received passive reporting, which was standard of care at the time of this study (ie, pharmacies had access to patient information but were not messaged directly).

The intervention began on January 1, 2017, defined as the index date, and lasted 4 months. The impact of the intervention on vaccination rates was assessed by conducting a retrospective data analysis based on the administrative claims data in the Humana Research Database (Louisville, Kentucky). Differences in the proportion of patients aged 65 to 89 years in intervention and control groups receiving a pneumococcal vaccination, as well as those aged 19 to 89 years receiving an influenza vaccination, during the 6 months post index (January 1, 2017, to June 30, 2017) were evaluated. A period of 6 months was used to capture the lag effect of the intervention on members who chose not to be vaccinated the day they were invited.

The study population for the pneumococcal vaccination analysis was composed of individuals aged 65 to 89 years with both medical and prescription drug coverage in the MAPD plan who filled prescriptions in participating pharmacies. The study population for the influenza vaccination analysis was composed of individuals aged 19 to 89 years with both medical and prescription drug coverage in the MAPD plan who filled prescriptions in participating pharmacies. Of note, the quality measure included patients as young as 6 months; however, there were no nonadult patients enrolled in this particular MAPD plan who filled prescriptions in participating pharmacies.

The analysis of pneumococcal vaccination in older adults included individuals who were pneumococcal vaccine naïve, aged 65 to 89 years, and continuously enrolled for 12 months preindex and 6 months post index, and had least 1 prescription fill between January 1, 2016, and December 31, 2016, at a participating pharmacy. Individuals with a documented pneumococcal vaccination (ie, PCV-13 or PPSV23)

 during the 12 months preindex were excluded from the pneumococcal measures, in accordance with the Consumer Assessment of the Healthcare Providers and Systems (CAHPS) survey.

 The look-back period to determine vaccination-naïve status was extended up to 5 years for individuals aged 65 to 69 years on January 1, 2017, preintervention, provided that the individual was enrolled with the health plan for that duration. For those who were not enrolled for 5 years preintervention, all available data were used (minimum 12 months). The influenza cohort included individuals without documentation of an influenza vaccine during July 1, 2016, to December 31, 2016; aged 19 to 89 years; and continuously enrolled for 12 months preindex and 6 months post index, with at least 1 prescription fill between January 1, 2016, and December 31, 2016, at a participating pharmacy. Individuals were excluded from both cohorts if they had a cumulative stay in a long-term care facility for at least 90 days during the study period.

The Humana Research Database, which contains claims data for individuals enrolled in Humana’s fully insured commercial and Medicare plans, was utilized for this study, along with data feeds from the participating community pharmacies. Patient enrollment data included information on demographics and coverage start and end dates. Medical claims data included information on physician visits, outpatient visits, hospital inpatient stays, and vaccinations administered in a practitioner’s office setting. Pharmacy claims data included information on prescription utilization, such as the specific medication filled, prescription fill date, quantity dispensed, and days’ supply.

The primary outcomes of interest were the proportion of individuals aged 65 to 89 years who received a pneumococcal vaccination and the proportion of individuals aged 19 to 89 years who received an influenza vaccination. The outcomes were calculated for the overall 6-month postindex time period.

Several additional variables were analyzed to compare the demographic and clinical characteristics of the intervention and control groups. The Deyo-Charlson Comorbidity Index was constructed from 

International Classification of Diseases, Tenth Revision, Clinical Modification

 diagnosis codes based on the methodology of Deyo and colleagues.16 Relevant codes were identified for the year prior to index (2016). Baseline health care resource utilization variables were constructed by separately counting the number of inpatient, emergency center, and outpatient visits for each patient during the year prior to the index date. The number of prescription fills during the preindex year was also tabulated for each patient. Patients who resided in Kentucky were assigned to the South region and patients from neighboring Indiana were assigned to the Midwest region based on US Census Bureau regional assignments. Age, gender, and race were derived from the enrollment files.

The baseline demographic and clinical characteristics were compared across intervention and control groups using descriptive statistics (ie, χ

 tests). Inverse probability of treatment weighting (IPTW) was used to adjust for baseline differences in the intervention and control groups. Standardized differences before and after adjustment were compared to evaluate balance on baseline patient characteristics in the weighted study cohorts, with a difference of 0.2 or less indicating small differences between the 2 groups.

 IPT-weighted hierarchical generalized linear models with a logit link and a binary distribution were constructed to examine the relationship of intervention status with immunization status. Individual pharmacies were included as a random effect to the models to account for potential clustering. Odds ratios (ORs) and 95% CIs were evaluated to determine statistical significance of relationships. The intraclass correlation coefficient was calculated, and the proportion of variability in the outcome accounted for by differences between the pharmacies, as opposed to differences among patients in the intervention and control groups, was reported. All analyses were performed using SAS Enterprise Guide Version 7.1.1 (SAS Institute).

After applying inclusion and exclusion criteria for the pneumococcal and influenza cohorts, the final sample sizes were 2135 and 2798, respectively (

). Of the individuals in the sample, 46.5% (n = 993) of the pneumococcal cohort and 44.1% (n = 1233) of the influenza cohort were attributed to intervention pharmacies. Upon comparison, the demographic characteristics of the intervention and control groups across the 2 cohorts exhibited similar patterns (

). The intervention and control groups in the pneumococcal older-adult cohort had similar ages, with a mean (SD) age of 73.0 (5.6) years for the intervention group and 73.4 (5.8) years for the control group (

= .1255). In the influenza cohort, the intervention group was older (67.7 [9.6] years) than the control group (66.4 [11.1] years) (

= .0006). The proportion of women in the total cohort was just more than half and did not differ significantly between intervention and control groups in the pneumococcal (

= .6836) cohorts. The geographic distribution was significantly different between intervention and control groups, likely due to the fact that the greater Louisville area (control group) included a portion of patients residing in southern Indiana (Midwestern region). The proportion of individuals qualifying for low-income subsidy status was lower in the intervention group vs the control group for both the pneumococcal cohort (16.6% vs 21.5%, respectively; 

= .0047) and the influenza cohort (22.7% vs 28.4%, respectively; 

A comparison of clinical characteristics indicated a few differences between the intervention and control groups for both immunization cohorts. The intervention group of both cohorts had a higher average number of prescriptions during the prior year than the control group (mean [SD], 25.0 [28.6] vs 18.3 [26.3] medications in the pneumococcal cohort; 

< .0001; and 33.5 [33.1] vs 22.9 [29.9] medications in the influenza cohort; 

< .0001). The Deyo-Charlson Comorbidity Index score was lower in the intervention group relative to the control group for both cohorts (mean [SD], 1.3 [1.8] vs 1.7 [2.2]; 

< .0001 for the pneumococcal cohort). The proportion of patients with specific comorbidities was similar between the intervention and control groups in the pneumococcal cohort, with the exception of diabetes, heart failure, and chronic obstructive pulmonary disease (COPD), where a lower proportion of the intervention group had these comorbidities. The pattern was similar in the influenza cohort. After applying IPTW, as can be seen in 

), standardized differences between the intervention and control groups generally decreased, and they indicated that the groups were well balanced, with only region having a difference greater than 0.2.

, during the 6-month postintervention period (January 1, 2017, to June 30, 2017), the proportion of older adults receiving pneumococcal vaccinations was higher in the intervention group (16.1%; n = 160) vs the control group (10.2%; n = 116) (

< .0001). In both the intervention group and control groups, approximately one-fourth of vaccinations were administered in the pharmacy and three-fourths were administered in the practitioner’s office. The number of patients receiving vaccinations declined each month from January until June for both groups.

Results for the generalized linear model estimating the impact of the intervention on pneumococcal vaccination in older adults are reported in 

. Unadjusted results indicated that patients in the intervention group had 1.70 times higher odds of receiving a pneumococcal vaccination than patients in the control group (

= .0126). After adjusting for demographic and clinical characteristics, the odds of receiving a vaccination were 1.91 times higher than the intervention group vs the control group (

= .0021). The effect of individual pharmacies accounted for 14% and 12% of the variation in the outcome in the unadjusted and adjusted models, respectively.

During the 6-month postintervention time period, the proportion of the influenza cohort receiving influenza vaccinations was higher in the intervention group (7.0%; n = 86) vs the control group (3.9%; n = 61) (

= .0003). In the intervention group, 41.9% (n = 36) of the vaccinations were received in a pharmacy and 58.1% (n = 85) of the vaccinations were received in the practitioner’s office, whereas for the control group only 31.1% (n = 19) of the vaccinations were received in store vs 68.9% (n = 42) received in the practitioner’s office. After IPTW was applied, as seen in 

, standardized differences between the 2 groups generally decreased and suggested that the groups were balanced, where only region and number of prescription fills showed differences greater than 0.2.

Results for the generalized linear model for the influenza vaccination are reported in 

. The unadjusted logistic regression results indicated that patients in the intervention group had 1.99 higher odds than the control group of receiving an influenza vaccination (

= .0030). After adjusting for demographic and clinical characteristics, the odds of receiving a vaccination were 2.18 times higher for the intervention group vs the control group (

= .0010). The presence of COPD increased the odds of receiving the influenza vaccination (OR, 1.67; 95% CI, 1.09-2.55). Individual pharmacies accounted for 11% of variability in the outcome in both models.

This quasi-experimental study found that a health plan–enabled pharmacist-led intervention was effective at increasing the odds that older adults aged 65 to 89 years receive the pneumococcal vaccination and the odds that individuals aged 19 to 89 years receive the influenza vaccination. Patients in the intervention group had twice the odds as patients in the control group of receiving a vaccination in both the pneumococcal and the influenza cohorts. These results suggest that pharmacists have a direct impact on immunization rates when provided with relevant data to support immunization gap closure.

The findings of this study are congruent with those of 3 systematic reviews, which found that pharmacy-based immunization programs increase the rate of adult immunizations.

 However, their impact varied widely, and a large number of the studies were not randomized. Another review reported on numerous case studies and observational analyses that demonstrated a positive impact of community pharmacists on immunization rates.

 However, this is 1 of only 2 studies of which we are aware that used a cluster randomized design to evaluate the impact of a health plan providing selective lists of patients to pharmacies.

Making the gaps in immunizations directly visible to the pharmacist may have contributed significantly to the success of this program. This relied on the completion of several critical steps. First, immunization gaps from administrative claims at the health plan were linked to pharmacies where the patients filled their prescriptions. Second, the gap information was shared with the specific community pharmacy where patients filled their prescriptions, via the EQuIPP platform. Third, once received, the pharmacist needed to confirm with patients that they had not received their immunization before offering it to them. Training ensured that the pharmacists understood these steps and acted upon them appropriately.

Pharmacist-led interventions may also be successful due to the convenience of visiting a pharmacy. Patients may visit a pharmacy to fill a prescription more frequently than they go to a physician’s office, and an individual’s pharmacy may be more conveniently located than the physician’s office. In addition, in-person contact may be preferred to a telephonic intervention, an example of which reported no impact.

 The acceptance of community pharmacists as immunizers is widespread, and all states allow pharmacists to provide at least some adult immunizations. In fact, more than 376,000 pharmacists have been certified as immunizers.

 Pharmacists are more geographically dispersed across Census tracts than are primary care physicians, increasing the availability of vaccine providers in areas with inadequate primary care provider coverage.

 Thus, a pharmacist-led in-person intervention has the potential to reach many more individuals than messaging from physician offices alone, regardless of where the patient chooses to receive the vaccination.

Notably, the number of prescription fills during baseline was significantly higher in the intervention group than in the control group for both the pneumococcal and influenza cohorts. This finding could imply that individuals in the intervention group frequented the pharmacy more regularly than individuals in the control group; thus, they may have had greater exposure to a pharmacist-led intervention, increasing the measured impact of the intervention. Nevertheless, the impact of the intervention remained significant even after adjusting for baseline differences.

Because this study used data from patients enrolled in 1 Medicare Advantage plan and was limited to pharmacies from 1 pharmacy chain in or near the state of Kentucky, generalization to the entire Medicare population should be done with caution. Additional intervention pilot programs in different regions, with different health plans and different retail pharmacies, would shed light on whether the impact of the intervention holds across alternative scenarios. Individuals filling their prescriptions via mail-order pharmacies would not be covered under this outreach plan. For these patients, the impact of telephonic or digital outreach, triggered by mail-order refills, might not be as high, because it would require initiative on the part of the patient to visit their pharmacy or physician’s office to receive a vaccination. The influence of temporal effects should also be evaluated in future studies by timing interventions during the fall months when vaccination rates tend to peak.

Considering the ongoing COVID-19 pandemic, the generalizability of findings in the current environment is unknown, because decreased access due to physical distancing and transportation reductions, as well as concerns about COVID-19 exposure, may hinder in-person pharmacist immunization programs. However, given the significant immunization service interruptions, opportunities for future health plan–pharmacy partnerships for COVID-19 vaccine administration and catch-up vaccinations that were missed during the pandemic should be considered.

Additionally, limitations common to studies using administrative claims data apply to this study. These include lack of comprehensive information regarding whether or not a patient received a pneumococcal or influenza vaccination. The database used was composed of claims submitted for payment. Therefore, any vaccinations received but not submitted for reimbursement were not captured. However, this intervention program enabled the pharmacist to learn directly from the patient whether the vaccinations were received. Lastly, for this study, individuals with either a PCV-13 or a PPSV23 vaccination at any time preindex (up to 5 years) were excluded from the pneumococcal measures in accordance with the CAHPS survey, which asked only whether patients had previously received a pneumococcal vaccination. However, guidelines at the time of the study recommended the PCV-13 followed 12 months later by the PPSV23 vaccination (note that more recently updated guidelines in 2019 recommend PPSV23, with shared decision-making used for PCV-13).

 As such, there may have been patients indicated for one of the pneumococcal vaccinations among those excluded according to the exclusion criteria.

This prospective pilot intervention study found that a pharmacist-led intervention was effective in increasing the number of older adults aged 65 to 89 years receiving the pneumococcal vaccination and individuals aged 19 to 89 years receiving the influenza vaccination. Patients in the intervention group were twice as likely as patients in the control group to receive a vaccination in both the pneumococcal older-adult cohort and the influenza cohort. Given the modest increase in absolute terms, however, continued efforts are needed to explore improvements in the vaccination rate. 

 Humana Healthcare Research Inc (RLS, MKP), Louisville, KY; Rx Quality Solutions (DPN), Ada, OH; Pharmacy Quality Solutions Inc (ND), Durham, NC; Rite Aid (JM), Camp Hill, PA; Premier Inc (ADB), Charlotte, NC; Pharmacy Quality Alliance (MP, PJC), Alexandria, VA.

 Pharmacy Quality Alliance received an unrestricted educational grant by Pfizer Inc to conduct the study.

 Mr Sheer and Dr Pasquale are employed by Humana Healthcare Research Inc, a wholly owned subsidiary of Humana Inc, that was paid by Pharmacy Quality Alliance, which received an unrestricted educational grant from Pfizer Inc to conduct this study. This study was conducted independent of health plan involvement. Dr Dorich is employed by Pharmacy Quality Solutions Inc, which operates the EQuIPP dashboard that was utilized in the study by pharmacists as the intervention tool. Dr Campbell has received grants from the Community Pharmacy Foundation, Merck, and Pfizer, and at the time of the study was employed by Pharmacy Quality Alliance, which received an unrestricted educational grant from Pfizer to conduct this study. Pharmacy Quality Alliance established Pharmacy Quality Solutions Inc as a for-profit, jointly owned company and holds a 51% share of the company. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (RS, DPN, ND, ADB, MP, MKP); acquisition of data (JM); analysis and interpretation of data (RS, DPN, MP, PJC, MKP); drafting of the manuscript (RS, DPN, MP, MKP); critical revision of the manuscript for important intellectual content (RS, ADB, MP, PJC, MKP); statistical analysis (RS); provision of patients or study materials (JM); obtaining funding (ADB, MP); administrative, technical, or logistic support (ND, JM, PJC, MKP); and supervision (ND, JM, MKP).

Richard L. Sheer, BA, Humana Healthcare Research Inc, 515 W Market St, Louisville, KY 40202. Email: rsheer1@humana.com.

1. Hak E, Nordin J, Wei F, et al. Influence of high-risk medical conditions on the effectiveness of influenza vaccination among elderly members of 3 large managed care organizations. 

2. Huang SS, Johnson KM, Ray GT, et al. Healthcare utilization and cost of pneumococcal disease in the United States. 

. 2011;29(18):3398-3412. doi:10.1016/j.vaccine.2011.02.088

3. 2020 topics & objectives: immunization and infectious diseases. HealthyPeople.gov. Accessed February 14, 2020. https://www.healthypeople.gov/2020/topics-objectives/topic/immunization-and-infectious-diseases/objectives

4. Thomas RE, Lorenzetti DL. Interventions to increase influenza vaccination rates of those 60 years and older in the community. 

. 2014;2014(7):CD005188. doi:10.1002/14651858.CD005188.pub3

5. Caffrey AR, DeAngelis JM, Ward KE, et al; Rhode Island Pharmacy Pneumococcal Vaccination Education Group. A pharmacist-driven academic detailing program to increase adult pneumococcal vaccination. 

. 2018;58(3):303-310. doi:10.1016/j.japh.2017.08.010

6. Van Amburgh JA, Waite NM, Hobson EH, Migden H. Improved influenza vaccination rates in a rural population as a result of a pharmacist-managed vaccination campaign. 

. 2001;21(9):1115-1122. doi:10.1592/phco.21.13.1115.34624

7. Krieger JW, Castorina JS, Walls ML, Weaver MR, Ciske S. Increasing influenza and pneumococcal vaccination rates: a randomized controlled study of a senior center–based intervention. 

. 2000;18(2):123-131. doi:10.1016/s0749-3797(99)00134-8

8. Isenor JE, Edwards NT, Alia TA, et al. Impact of pharmacists as immunizers on vaccination rates: a systematic review and meta-analysis. 

. 2016;34(47):5708-5723. doi:10.1016/j.vaccine.2016.08.085

9. Baroy J, Chung D, Frisch R, Apgar D, Slack MK. The impact of pharmacist immunization programs on adult immunization rates: a systematic review and meta-analysis

. 2016;56(4):418-426. doi:10.1016/j.japh.2016.03.006

10. Beal JL, Kadakia NN, Reed JB, Illingworth Plake KS. Pharmacists’ impact on older adults’ access to vaccines in the United States. 

. 2020;38(11):2456-2465. doi:10.1016/j.vaccine.2020.01.061

11. Pneumococcal vaccination among U.S. Medicare beneficiaries aged ≥65 years, 2009-2017. CDC. Accessed December 18, 2020. https://www.cdc.gov/vaccines/imz-managers/coverage/adultvaxview/pubs-resources/pcv13-medicare-beneficiaries.html

12. Chepigin C, Adelsberg M, Hughes R IV, Kane R. A comprehensive analysis of influenza vaccine uptake in Medicare. Poster presented at: National Adult and Influenza Immunization Summit; May 14-16, 2019; Atlanta, GA. Accessed December 11, 2020. https://www.izsummitpartners.org/content/uploads/2019/05/
1-comprehensive-analysis-of-flavax-uptake-in-medicare.pdf

13. Tomczyk S, Bennett NM, Stoecker C, et al; CDC. Use of 13-valent pneumococcal conjugate vaccine and 23-valent pneumococcal polysaccharide vaccine among adults aged ≥65 years: recommendations of the Advisory Committee on Vaccination Practices (ACIP). 

14. CDC. Use of 13-valent pneumococcal conjugate vaccine and 23-valent pneumococcal polysaccharide vaccine for adults with immunocompromising conditions: recommendations of the Advisory Committee on Vaccination Practices (ACIP). 

15. Consumer Assessment of Healthcare Providers & Systems (CAHPS). CMS. Accessed February 14, 2020. https://www.cms.gov/Research-Statistics-Data-and-Systems/Research/CAHPS/index.html

16. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with 

. 1992;45(6):613-619. doi:10.1016/0895-4356(92)90133-8

17. Faraone SV. Interpreting estimates of treatment effects: implications for managed care. 

Statistical Power Analysis for the Behavioral Sciences

19. Burson RC, Buttenheim AM, Armstrong A, Feemster KA. Community pharmacies as sites of adult vaccination: a systematic review. 

. 2016;12(12):3146-3159. doi:10.1080/21645515.2016.1215393

20. Stolpe S, Choudhry NK. Effects of automated immunization registry-based telephonic interventions on adult vaccination rates in community pharmacies: a randomized controlled trial. 

. 2019;25(9):989-994. doi:10.18553/jmcp.2019.25.9.989

21. Who we are. American Pharmacists Association. Accessed September 16, 2021. https://pharmacist.com/who-we-are

22. Shah PD, Trogdon JG, Golden SD, Golin CE, Marciniak MW, Brewer NT. Impact of pharmacists on access to vaccine providers: a geospatial analysis. 

. 2018;96(3):568-592. doi:10.1111/1468-0009.12342

23. Matanock A, Lee G, Gierke R, Kobayashi M, Leidner A, Pilishvili T. Use of 13-valent pneumococcal conjugate vaccine and 23-valent pneumococcal polysaccharide vaccine among adults aged ≥65 years: updated recommendations of the Advisory Committee on Immunization Practices. 

. 2019;68(46):1069-1075. doi:10.15585/mmwr.mm6846a5

Articles

Patients whose pharmacy receives notification of their immunization gap have twice the odds of receiving immunizations compared with those whose pharmacy does not receive the notification.

Medicare Advantage–Pharmacy Partnership Improves Influenza and Pneumococcal Vaccination Rates

 To develop and validate predictive models for imminent fracture risk in a Medicare population.

 This retrospective administrative claims (Humana Research Database) study assessed imminent risk in Humana’s Medicare Advantage and Prescription Drug plan members.

 Individuals (aged 67-87 years on January 1, 2015 [index]) with 1 year or more of history were followed for 3 months to up to 2 years, with censoring at death/disenrollment. The cohort was split into training and validation samples (1:1). Cox regression models assessed demographics, fracture history, medically significant falls, osteoporosis-related factors, frailty markers, and selected medications and comorbidities for independent predictors (

 <.001) of incident nontraumatic clinical fractures in 12 and 24 months. A 6-variable model of 12-month risk used a published method for the risk-scoring point system.

 Of 1,287,354 individuals (mean age, 74.3 years; 56% female; 84% white), 3.8% had at least 1 fragility fracture at 12-month follow-up; 6.6% experienced fracture at 24 months (women vs men: 12 months, 4.8% vs 2.5%; 24 months, 8.3% vs 4.4%; both 

 <.01). At 12 months, recent fracture conferred approximately 3-fold-higher fracture risk (vs no recent fracture). Older age, white race, female sex, osteoporosis-related screening/diagnosis/medication, antidepressant/antipsychotic/sedative hypnotic/muscle relaxant medications, history of falls, fracture history, and respiratory conditions also increased risk (all 

 <.0001). The simplified model (recent fracture, age, sex, race, falls, antidepressant/antipsychotic/sedative hypnotic/muscle relaxant medications) performed well (C statistic = 0.71).

 Recent fracture, older age, female sex, white race, falls, and antidepressant/antipsychotic/sedative hypnotic/muscle relaxant medications predict imminent fracture risk in an older-adult Medicare Advantage population. Imminent fracture risk can be assessed using 6 easily quantified factors.

Guideline-recommended tools that estimate long-term (10-year) fracture risk are used for osteoporosis treatment decisions, but these tools do not quantify imminent (in the next 1-2 years) risk.

In our validated predictive models for imminent (1-year) fracture risk in the elderly population, age, race, sex, osteoporosis-related screening/diagnosis, medically significant falls, antidepressant/antipsychotic/sedative hypnotic/muscle relaxant medications, history of fracture, respiratory disorders, nervous system disorders, physical factors, and additional comorbidities/medications were significantly associated with greater imminent fracture risk.

A simplified model (recent fracture, age, sex, race, falls, and antidepressant/antipsychotic/sedative hypnotic/muscle relaxant medications) supports feasibility for hand-calculating imminent risk during routine visits.

Up to 25% of men and 50% of women will experience a fracture in their remaining lifetime after age 50 years.

 Osteoporosis-related fractures confer significant direct medical costs, quality-of-life decrements and functional disability, and indirect costs (lost productivity and caregiver burden).

 By 2025, fracture-related direct treatment costs are expected to reach $25.3 billion in the United States.

 The impact of fractures to individuals may be long-lasting and result in increased use of healthcare services, a greater likelihood of hospitalization and skilled nursing care, reduced economic circumstances, and increased mortality risk for up to 10 years post fracture.

Various antiresorptive and bone-forming therapies have received FDA approval based on their ability to reduce the risk of fracture in men and women with osteoporosis.

 Fracture risk-assessment tools help clinicians estimate the risk of osteoporosis-related fracture in their patients, and these estimates inform treatment recommendations. The most commonly used tools, such as FRAX,

 focus exclusively on long-term fracture risk (ie, 10-year time horizon). The literature suggests that the imminent risk (ie, in next 1-2 years) of fracture also has important implications for treatment decisions.

 Imminent risk is responsive to changes in bone mineral density (BMD), fracture, and fall history, as well as factors that influence physical or cognitive functioning in ways that are not captured in existing tools.

 Furthermore, patients with osteoporosis may not perceive fracture risk accurately, and long-term risk assessments, in conjunction with patients’ potentially incomplete understanding of the relationship between osteoporosis and fracture risk, may not provide sufficient motivation for treatment initiation.

Elevated imminent risk may require a different treatment approach from fracture risk that accumulates or extends over a longer time frame.

 The objective of this study was to develop and validate predictive models for imminent risk of fracture in a population of older adults enrolled in Humana’s Medicare Advantage and Prescription Drug (MAPD) plans. This paper describes the primary full model, which includes an extensive set of predictors for the outcome of any fracture within a 12-month follow-up, and a parsimonious model, which was developed to provide proof of concept for a risk-assessment tool that could easily be used for hand-calculating risk during a routine physician office visit.

This retrospective database study used a cohort design and multivariate analysis to identify predictors of imminent fracture risk in older adults enrolled in Humana MAPD plans. The Humana Research Database (Humana; Louisville, Kentucky), which contains administrative data for Humana’s fully insured commercial and Medicare individuals, was used. Data on MAPD enrollment, medical claims, and outpatient pharmacy claims were linked for each individual. The study period—January 1, 2014, through December 31, 2016&mdash;was split into a 1-year baseline period (January 1, 2014, through December 31, 2014) and a 2-year follow-up period (January 1, 2015, through December 31, 2016).

All study individuals were aged 67 to 87 years on the index date (January 1, 2015) and had continuous enrollment in an MAPD plan throughout the baseline period and for at least the first 3 months of the follow-up period. Follow-up was censored at the date of death or plan disenrollment for individuals lost to follow-up before the end of study. Individuals with an 

International Classification of Diseases, Ninth Revision, Clinical Modification

) diagnosis of either nonmelanomatous cancer or Paget disease of the bone and individuals enrolled in administrative services—only health plans or in a Humana health plan which is contractually excluded from research studies were excluded.

The sample was split into a randomly generated training (development) sample and a validation sample, with a 1:1 ratio. The primary study outcome was defined as incident fragility fracture. These fractures were identified using a combination of 

 diagnosis codes. Clinical fragility (nontraumatic) fractures with an osteoporosis likelihood score of 8 or greater, as defined by Warriner et al,

 were identified during 12- and 24-month follow-up intervals. This published algorithm was developed to enhance the accuracy of identifying osteoporosis-related fractures in administrative claims data and was used in the current study to reduce the potential for misclassification of the primary study outcome. Fracture codes recorded during the follow-up period but within 60 days of a baseline fracture were considered as a continuation of care for the baseline fracture and, therefore, not considered in the analysis of fracture outcomes during follow-up.

Covariates considered as potential independent variables for the predictive models included patient demographics (age, sex, race, and geographic region), history of fracture, falls that result in medical encounters or that require treatment (ie, medically significant falls), BMD testing, comorbidities, medication use, and markers of frailty. Comorbidities were quantified by the Deyo-Charlson Comorbidity Index

 score and yes/no indicator variables for selected comorbidities, including those associated with an increased likelihood of fractures/medically significant falls. Osteoporosis treatments, as well as medications associated with an increased likelihood of fractures/medically significant falls (eg, antihypertensives, anticonvulsants, antidepressants, antipsychotics), were assessed. Markers of frailty included data on difficulty walking, use of durable medical equipment or ambulance/life support, paralysis, weakness, and podiatric care.

The training sample was used to construct risk-prediction models for any fragility fracture occurring in the 12- and 24-month follow-up intervals. The primary analysis focused on incident fractures that occurred in the 12-month follow-up, and data from the 24-month follow-up provided a sensitivity analysis. Here we present predictive models for any fragility fracture in these 2 follow-up periods, as well as results from a model specific to hip fracture risk during a 12-month follow-up.

To accommodate variable length of follow-up for individuals in the training and validation samples, Cox regression models

 were used to estimate adjusted hazard ratios (HRs) for potential predictor variables. Variables significantly associated with fracture risk (

 <.001), after controlling for all other predictors in the model, were considered to be independent predictors. Using the validation sample, the predictor coefficients from the finalized predictive models of fracture risk for hip fracture were applied to the patient baseline characteristics to obtain the likelihood of fracture risk for each patient. C statistics and other measures (false-positive rate, false-negative rate, sensitivity, specificity, positive predictive value, negative predictive value, and efficiency [ie, percentage of accurately predicted patients in the study sample]) were computed to assess model performance when data from the validation sample were used. Models are typically considered effective when the C statistic is greater than 0.7 and strong when it exceeds 0.8.

 Therefore, a minimum threshold of 0.7 for the C statistic was considered when comparing the models. Additional performance metrics were assessed at fracture probabilities of 5% or greater, 10% or greater, 15% or greater, 20% or greater, and 25% or greater.

A parsimonious version of the primary predictive model was developed to adapt the results of the analysis for use as an imminent (12-month) fracture risk assessment for an individual patient in the routine care setting. The goal was to create a parsimonious model with approximately 5 variables and to incorporate a risk-scoring point system based on the Sullivan methodology

 adapted for use with 1-year incidence estimates. In the parsimonious model, increasing numbers of points were assigned to increasing levels of risk (eg, increasing age was associated with a greater number of points), and the totaled points corresponded to an individual’s predicted fracture probability during 12 months of follow-up.

From Humana’s MAPD population, 1,287,354 individuals (mean age, 74.3 years; 56% female; 84% white race) met all criteria for inclusion in this study (

). Only 6% of these individuals had a history of falls during the baseline period. Approximately 4% of the individuals identified had a claim for a recent fracture (ie, within the baseline year) and fewer than 1% had a historical fracture at any time in recorded medical history. Although 19% of individuals had a diagnosis of osteoporosis, prescription fills for osteoporosis medications during the baseline period were relatively uncommon (6%). By contrast, 72% of individuals had been given a diagnosis of hypertension, and 64% of the total population used antihypertensives. Approximately 15% to 20% of individuals had used other medications of interest.

In the 12-month follow-up period, 3.8% of individuals experienced at least 1 fragility fracture, and the fracture incidence increased to 6.6% over the 24-month follow-up period (

). Fracture incidence was higher among women than men over both follow-up intervals (12 months, 4.8% vs 2.5%; 24 months, 8.3% vs 4.4%; 

 <.01 for both). The incidence of hip fractures in men and women combined was 0.7% in the 12-month follow-up period and 1.4% in the 24-month follow-up period.

In the primary analysis, recent fracture (ie, within the past year prior to the start of the follow-up period) had the strongest association with 1-year fracture risk (

); this factor conferred nearly a 3-fold-higher risk among the affected individuals compared with the fracture risk in individuals who had no recent fracture. Risk also increased steadily with age and was highest among individuals aged 85 to 87 years. White race and female sex were also strongly associated with increased fracture risk, as were osteoporosis-related screening and diagnosis, the use of osteoporosis-related medications, use of antidepressant/antipsychotic/sedative hypnotic/muscle relaxant medications, falls, history of fracture, and respiratory conditions (all 

 <.0001). The primary model performed well, with a C statistic of 0.73, false-positive rates ranging from 74% to 90%, and false-negative rates ranging from 2% to 4% across fracture probability cut-offs ranging from 5% or greater to 25% or greater (

 [A]). Across these same fracture probability categories, the model’s specificity was better than its sensitivity. Higher areas under the curve and C statistics were observed with the 12-month model compared with the 24-month model developed as a sensitivity analysis (eAppendix Table 1 [B]; eAppendix Figure 2 [B]). Risk factors for hip fracture within 12 months exhibited significant overlap with the predictors identified for all fragility fractures during that time interval (Table 3). In particular, older age and recent fracture exerted the greatest influence on risk, with 2- to 3-fold increases in risk associated with these factors. The C statistic for the 12-month hip fracture model was 0.77.

The parsimonious model for any fracture, which included recent fracture, age, sex, race, falls, and antidepressant/antipsychotic/sedative hypnotic/muscle relaxant medications as predictors, exhibited good performance with a C statistic of 0.71 (

). The scored points and corresponding predicted probability of fracture were compared with actual fracture rates in 2015 for individuals with each point total (

). These rates increased as the number of points increased.

This study identified predictors of the imminent risk of fracture (ie, within 12-24 months) among older adults enrolled in Humana’s MAPD plans. Of the 75 patient characteristics initially considered for the predictive models, recent fracture (ie, fracture occurring during the 1-year baseline period), advanced age, and white race were most strongly associated with increased imminent risk of fractures. Other key factors included female sex, indicators of osteoporosis (screening, diagnoses, and medications), any history of fracture, falls, and antidepressant/antipsychotic/sedative hypnotic/muscle relaxant medications.

The risk factors identified in this study are consistent with those identified in prior observational studies that assessed imminent fracture risk in different populations, including individuals with and without osteoporosis. In osteoporosis, generally, white/Caucasian women are at greater risk of fracture than nonwhite women or men of similar age.

 Low BMD, prior fracture, and falls are also well-documented risk factors for fragility fracture.

 The growing body of research on risk factors for imminent fractures indicates that these factors, along with certain comorbidities and medications that affect physical and cognitive functioning, also influence the risk of fracture in the near term.

 The model results for hip fracture illustrate this last point well, with previous research indicating that various mechanisms may influence the association between dementia and hip fractures, including common risk factors for both and the role of dementia and its treatment as an indirect risk factor that increases fracture risk through intermediate risk factors (eg, falls, vitamin D deficiency, treatment adverse effects).

We found that antidepressant/antipsychotic/sedative hypnotic/muscle relaxant medications had a significant impact on the imminent risk of fracture. It is possible, as recent literature suggests, that these medications increase the risk of falls, thereby indirectly increasing the risk for fragility fractures.

 Although the 2 variables—antidepressant/antipsychotic/sedative hypnotic/muscle relaxant medications and falls&mdash;were not highly correlated (Pearson correlation coefficient = 0.13), this may be because falls are vastly underreported in medical claims data, which would make it difficult to show this association in the available data.

The accumulating literature on short-term or imminent risk highlights the fact that fracture risk is not linear but varies over time as a patient’s characteristics and exposures change, and investigators have begun identifying factors that contribute specifically to increased risk within the next 1 to 2 years.

 An important aspect of several risk factors identified in our full models of any fracture and in previous studies of imminent risk is that they are dynamic and their influence on fracture risk occurs quickly. Prior fracture illustrates this concept well. There is strong evidence that the risk of fragility fracture is approximately doubled after an initial fracture and that the greatest increase in risk occurs within the first year post fracture.

 Our results confirm this pattern of risk by demonstrating that although any history of fracture is a significant predictor of imminent fracture risk (adjusted HR, 1.26; 95% CI, 1.17-1.36), the impact of recent fracture is more than twice as great (adjusted HR, 2.81; 95% CI, 2.70-2.92).

Fracture risk may be particularly variable in older individuals because events that we have identified as predictors of imminent risk, such as the occurrence of a stroke or diagnosis of a neurologic or psychologic disorder (eg, Alzheimer disease, depression, anxiety), generally become more common with age. Similarly, changes in physical or cognitive functioning, which may be related to the onset of new comorbidities or the initiation of new medications, may either increase fracture risk directly or act on fracture risk by increasing a patient’s risk of falls as mentioned previously.

 This inherent malleability of fracture risk underscores the importance of being able to quantify imminent risk as a patient’s medical status changes and to help patients understand the need for timely intervention to reduce the risk of fracture and its related sequelae.

This dynamic nature of fracture risk suggests a need for tools that can be used to regularly assess the changing risk of imminent fracture in older adults. In this population, such risk assessments should be done regularly, and more frequent use of a tool such as the one developed in this study might be useful for patients whose health status is in transition or particularly modifiable. Quantifying imminent risk in an ongoing, real-time fashion requires a simple risk-assessment tool. With this in mind, we developed a parsimonious model based on our full model and adapted this model into a prototype for a points-based risk-assessment tool. This model uses only 6 variables (recent fracture, age, sex, race, falls, and antidepressant/antipsychotic/sedative hypnotic/muscle relaxant medications) and was adapted to a point-based system, with total points (risk score) easily computed during a routine physician office visit. Given the importance of falls as a predictor of long-term fracture risk and their association with imminent risk of fracture, additional parsimonious models were tested to assess a composite physical factors variable as a potential predictor of imminent risk. Because physical factors may influence fall risk, in one model, the physical factor variable was simply substituted for falls in the existing parsimonious model, whereas in a second model, falls and physical factors were combined and tested with age, sex, race, and recent fracture. The model performance was similar in both of these versions and resulted in the same number of points being attributed to this fifth variable.

This work supports the feasibility of a formal risk assessment that can quickly and easily be used during physician office visits to identify patients who are at high risk for an imminent fracture. To guide the use of this tool, input could be obtained from physician focus groups to help determine clinically meaningful point score thresholds for fracture prevention initiatives. Such initiatives would aid in the identification of patients in the greatest need of timely osteoporosis management ranging from BMD screening to interventions including prescription medications and initiatives to reduce the risk of falls.

&#8203;&#8203;&#8203;&#8203;&#8203;&#8203;

This study is subject to limitations common to administrative claims data research. Identifying fragility fractures and distinguishing incident fractures from prevalent fractures can be challenging with the information available in claims. To address this issue, we excluded all fractures that were recorded as trauma related and examined the timing of fracture-related care to identify and exclude claims most likely to represent care for prevalent fractures that occurred in the baseline period. We also limited the analysis to closed fractures with the highest likelihood of being classified as fragility fractures based on a previously published methodology designed for use with administrative claims data.

 Specifically, we included only fractures with an osteoporosis likelihood score of 8 or greater to reduce bias related to potential misclassification of fracture outcomes. There was also potential for selection bias, given the requirement for at least 12 months of continuous insurance coverage in the baseline period. Although the multivariate regression modeling used in this study would have reduced the potential for bias among measured covariates, bias associated with unmeasured covariates would have remained. We also note that results may not generalize to the full US population, as the geographic mix in the sample was not balanced across all regions.

In summary, recent fracture, older age, female sex, white race, falls, and use of antidepressant/antipsychotic/sedative hypnotic/muscle relaxant medications were strong predictors of the imminent risk of fracture in Medicare-enrolled older adults. A patient’s level of imminent risk may change over time, and this work provides proof of concept that the imminent risk of fracture can be assessed by evaluating a small number of factors. Administering such a risk-assessment tool regularly may improve physicians’ opportunities to identify elderly patients at high risk of imminent fractures and enhance their ability to provide timely interventions (screening, medications, and fall-prevention strategies). Additional research is needed to set clinically meaningful risk-score thresholds for intervention, determine the applicability and accuracy of this risk-score model to specific patient subgroups, and evaluate the best assessment schedule and optimal ways to implement this tool within the existing physician workflow.&ensp;

The authors would like to thank Sally Wade (Wade Outcomes Research and Consulting) for medical writing assistance.

 Humana Healthcare Research, Inc, Humana Inc (RLS, LS, MKP), Louisville, KY; Amgen Inc (RLB), Thousand Oaks, CA.

 Mr Sheer is an employee of and stockholder in Humana Inc. Mr Barron was employed at Amgen until May 2018 and owned stock in Amgen as an employee. Ms Sudharshan was employed at Humana at the time of this study; Humana received consulting fees from Amgen for the conduct of this study. Dr Pasquale owns stock in Amgen and is an employee of Humana Health Research, which received funding from Amgen to conduct the study.

 Concept and design (RLS, RLB, LS, MKP); analysis and interpretation of data (RLS, RLB, LS, MKP); drafting of the manuscript (RLS, MKP); critical revision of the manuscript for important intellectual content (RLS, RLB, LS, MKP); statistical analysis (RLS, LS); provision of patients or study materials (LS); obtaining funding (RLB); administrative, technical, or logistic support (RLB, LS); and supervision (RLB, MKP).

 Margaret K. Pasquale, PhD, Humana Healthcare Research, Inc, Humana Inc, 500 W Main St, Louisville, KY 40202. Email: mpasquale@humana.com.

1. What is osteoporosis and what causes it? National Osteoporosis Foundation website. nof.org/patients/what-is-osteoporosis. Accessed April 3, 2018.

2. Burge R, Dawson-Hughes B, Solomon DH, Wong JB, King A, Tosteson A. Incidence and economic burden of osteoporosis-related fractures in the United States, 2005-2025. 

. 2007;22(3):465-475. doi: 10.1359/jbmr.061113.

3. Dyer SM, Crotty M, Fairhall N, et al; Fragility Fracture Network (FFN) Rehabilitation Research Special Interest Group. A critical review of the long-term disability outcomes following hip fracture. 

. 2016;16:158. doi: 10.1186/s12877-016-0332-0.

4. Fischer S, Kapinos KA, Mulcahy A, Pinto L, Hayden O, Barron R. Estimating the long-term functional burden of osteoporosis-related fractures. 

This study provides proof of concept that imminent risk of fracture can be assessed by evaluating recent fracture, age, sex, race, medically significant falls, and psychoactive medications.

Richard L. Sheer, BA

Articles

Medicare Advantage–Pharmacy Partnership Improves Influenza and Pneumococcal Vaccination Rates

Validated Prediction of Imminent Risk of Fracture for Older Adults