Protocol for the Models of Primary Osteoporosis Screening in Men (MOPS) Cluster Randomized Trial

Abstract

Current guidelines recommend primary osteoporosis screening for at-risk men to reduce the morbidity, mortality, and cost associated with osteoporotic fractures. However, analyses in a national Veterans Health Administration cohort of over 4,000,000 men demonstrated that primary osteoporosis screening as it is currently operationalized does not benefit most older Veterans due to inefficient targeting and low subsequent treatment and adherence rates. The overall objective of this study is to determine whether a new model of primary osteoporosis screening reduces fracture risk compared to usual care. We are conducting a pragmatic group randomized trial of 38 primary care teams assigned to usual care or a Bone Health Service (BHS) screening model in which screening and adherence activities are managed by a centralized expert team. The study will: 1) compare the impact of the BHS model on patient-level outcomes strongly associated with fracture rates (eligible proportion screened, proportion meeting treatment criteria who receive osteoporosis medications, medication adherence, and femoral neck bone mineral density); 2) quantify the impact on provider and facility-level outcomes including change in DXA volume, change in metabolic bone disease clinic volume, and PACT provider time and satisfaction; and 3) estimate the impact on health system and policy outcomes using Markov models of screening program cost per quality adjusted life year based from health system and societal perspectives.

Introduction

One in five men over age 50 will suffer a major fracture in their remaining lifetime.¹ Among patients who fracture, men are more than twice as likely to experience complications including mortality.^2,3 Fractures are associated with functional decline and poor quality of life.² Hip fractures alone cost $43 million to the Veterans Health Administration (VA) annually.⁴ Fortunately, osteoporotic fractures are preventable. Screening involves a non-invasive, inexpensive Dual Energy X-ray Absorptiometry (DXA) measurement of bone mineral density (BMD).^{1, 5} Once high-risk patients are identified, first line medications such as alendronate (annual VA cost $20), reduce the risk of vertebral, hip and other fractures by up to 50%^{6 7–9} while also decreasing mortality.^{10, 11}

However, men are rarely treated for osteoporosis before a fracture has occurred.^12–14 A national study of primary osteoporosis screening in male Veterans showed screening rates of only 8% for men over age 65; far lower than expected based on the prevalence of osteoporosis risk factors. Even among men in whom screening was completed, it was not associated with lower overall fracture rates because osteoporosis treatment and adherence following screening were extremely low. Nearly half of men who met treatment criteria were never prescribed medication, and among those who were, 90% discontinued it within 1 year.¹⁵ Available data suggests that this problem is likely even worse for non-Veteran men.

Attempts to improve osteoporosis screening using quality improvement programs have been minimally effective.^16–18 Electronic Health Record (EHR) alerts do not improve screening rates.¹⁹ However, two distinct osteoporosis screening paradigms have been suggested. First, a practice manager paradigm in which a non-physician staff member manages screening led to a doubling in DXA orders in a single academic medical center.¹⁹ In this approach, responsibility remains at the practice level. In contrast, a Fracture Liaison Service (referred to here as “Bone Health Service”, BHS) represents a centralized model that has been successful in improving secondary osteoporosis screening and treatment adherence after a fracture has already occurred.^{20, 21} In this model, a team including nurses and a bone specialist identify all patients with fracture within a health system and arrange for evaluation and treatment. Such models have reduced 2-year fracture rates by 56%²² and are highly cost-effective.^{23, 24} The National Bone Health Alliance recommends expansion of BHS models for high risk individual prior to their first fracture, however BHS’ impact on fracture risk and its cost-effectiveness for primary prevention is unknown.

Attempts to improve osteoporosis medication adherence have also had mixed success. A systematic review found evidence for interventions which provided frequent telephone follow-up.²⁵ Patient education on the risks and benefits for treatment is another key component across multiple studies.²⁶ Less evidence is available for more efficient adherence interventions used in other conditions, such as text reminders²⁷ or patient tools to identify reasons for nonadherence.²⁸

Based on this body of evidence, we designed a group randomized trial to measure the impact and cost-effectiveness of novel models of primary osteoporosis screening in older male Veterans at high risk for fracture.

Methods

Conceptual Frameworks

Methods were designed using the Population-based Research Optimizing Screening through Personalized Regimens (PROSPR)²⁹ model, adapted for osteoporosis screening (Figure 1). Because osteoporosis medication adherence is a critical component of screening effectiveness, we have also adopted the Patient Preference Adherence Model³⁰ (Figure 2) to address patient, provider, and system-level barriers previously reported for osteoporosis medications.^{26, 31–35}

Figure 1.

Modified PROSPR Model for Osteoporosis Screening for Usual Care and Bone Health Services (BHS) Arms, with Study Outcomes.

Figure 2.

Conceptual Model of Medication Adherence Factors.

Design.

Pragmatic group randomized trial of male Veterans aged 65–85 years meeting VA Undersecretary screening guidelines. Primary care teams (n=39 teams) were initially randomized into 3 groups: usual care (control); a centralized Bone Health Service (BHS) model; or a Practice Management model. In the Practice Management Model, teams were to be provided with a prioritized list of high-risk Veterans meeting screening guidelines quarterly, an osteoporosis medication adherence report, and additional tools and resources to assist with osteoporosis screening and adherence monitoring such as electronic order sets and patient education materials. However, shortly after randomization the study was suspended due to the COVID-19 pandemic. When the study was approved to resume in September 2020, it became clear from discussions with enrolled teams that the workflow changes caused by the pandemic made the Practice Management arm impracticable. With permission from the funding agency, teams initially randomized to the Practice Management arm were re-randomized to either Usual Care or BHS, and the design was modified from a 3-arm to a 2-arm group randomized trial. One team declined re-randomization due to staff turnover, resulting in n=38 teams.

Outcomes for up to 100 Veterans at highest risk for fracture within each randomized primary care team will be assessed by investigators blinded to group assignment via EHR at baseline and 2 years. The 2-year follow-up period allows for enough time to offer screening of at least 100 eligible men within each team and for long-term osteoporosis medication adherence and change in BMD to be measured. Intention to treat analysis will be employed.

All study activities are approved by the Durham and Richmond VA Institutional Review Boards, a local Safety Committee comprised of 3 clinician researchers outside of the study team, and the VA Health Services Research and Development Data and Safety Monitoring Board. Clinical Trials registration number NCT04079868.

Setting.

Non-specialty primary care teams within 2 VA Health Systems in North Carolina and Virginia. These include teams based in academically affiliated hospitals, community health centers, and rural outpatient clinics.

Subjects.

Two levels of subjects are considered. Primary Care Providers (n=76) include the medical provider (MD or advanced practice provider, APP) and registered nurse (RN) who provide primary care to a panel of 800–1000 patients. Women’s Health and specialty primary care teams serving younger Veterans are excluded. Patients (estimated n=2280–3800) include men aged 65–85 years eligible for primary osteoporosis screening within enrolled primary care teams. Inclusion criteria include no prior fracture or osteoporosis diagnosis, and the presence of at least 1 fracture risk factor as specified in current VA guidelines (weight loss >20% in 5 years; BMI <25  kg/m²; diabetes; pernicious anemia; gastrectomy; anticonvulsants; glucocorticoids; androgen deprivation therapy; hyperthyroidism; hyperparathyroidism; rheumatoid arthritis; alcohol dependence; chronic lung disease; chronic liver disease; stroke; Parkinsonism; prostate cancer; or current smoking).³⁶ Exclusion criteria include enrollment in hospice or palliative care. If more than 100 men in a primary care team meet inclusion criteria at the time of team initiation, they are further risk stratified using the Osteoporosis Screening Tool (OST)³⁷ and the 100 highest risk patients are included. Baseline patient characteristics are in Table 1.

Table 1.

Baseline characteristics of primary care teams (n=39) and Veterans eligible for screening (n= 13,026) randomized in this study.

Primary Care Teams
Characteristic	Usual Care n=19	Bone Health Service n=20	P value
Location
Academic Medical Center	10	11
Community Clinic	9	9
Provider type
MD	15	15
Advanced Practice Provider	4	5
Patients Eligible for Screening
Characteristic	Usual Care n= 6321	Bone Health Service n=6705	P value
Mean age yrs (SD)	69.1 (7.0)	70.1 (7.4)	<0.001
Mean BMI (kg/m2) (SD)	30.1 (6.0)	30.0 (5.9)	0.38
Race (%)
White	3342 (52.8)	3651 (54.4)	0.07
Black	2803 (44.3)	2869 (42.7)	0.07
Other/Unknown	176 (2.7)	185 (2.7)	NS
Chronic lung disease (%)	891 (14.0)	1056 (15.7)	0.006
Diabetes (%)	2265 (35.8)	2571 (38.3)	0.003
Hyperthyroidism	14 (0.2)	28 (0.4)	0.04
Hyperparathyroidism (%)	39 (0.6)	32 (0.4)	0.11
Parkinson’s (%)	99 (1.5)	101 (1.5)	NS
Prostate cancer (%)	568 (9.0)	642 (9.6)	0.24
Rheumatoid arthritis (%)	107 (1.7)	90 (1.3)	0.15
Gastrectomy or malabsorption (%)	129 (1.9)	163 (2.4)	0.05
Alcohol abuse (%)	678 (10.7)	655 (9.7)	0.06
Chronic liver disease (%)	360 (5.6)	329 (4.9)	0.07
Chronic kidney disease stage 4 or 5 (%)	80 (1.2)	61 (0.9)	0.09
Smoking (%)	1310 (20.7)	1229 (18.3)	<0.001
Androgen deprivation therapy (%)	56 (0.8)	55 (0.8)	NS
Glucocorticoids (%)	1435 (22.7)	1586 (23.6)	0.22
Traditional anti-epileptic (%)	1346 (21.2)	1444 (21.5)	0.68
Median OST* score (range)	4.50 (−14.7–27.0)	4.30 (−8.9–35.0)	0.55

^*Osteoporosis Self-Assessment Tool = [Weight (kg) – Age (years)]/5. Score ≤ −3 indicates high risk, score >−3 and <1 indicates moderate risk, ≥ 1 indicates low risk

Recruitment.

Study procedures were described to primary care teams during a meeting. Study personnel followed-up individually with each provider for informed consent; a waiver of documentation was obtained from the site IRBs. Each primary care team consists of a single medical provider, Registered Nurse (RN), Licensed Practical Nurse (LPN) and medical assistant. RNs occasionally work with more than one provider; to prevent contamination, only the first provider expressing interest in the study was eligible. A waiver of patient informed consent and HIPAA authorization was granted for patient outcomes assessment via the EHR because this is a minimal risk study implementing routine clinical care as currently recommended by the VA Undersecretary.

Randomization.

Stratified, block randomization at the level of the primary care team was employed. Teams were stratified first by practice setting (hospital based, community clinic) and then medical provider type (MD vs. Advanced Practice Provider) and randomized in blocks by a statistician unaware of team identity using SAS 9.4.

Interventions.

The BHS intervention includes two separate components; 1) osteoporosis screening promotion, and 2) medication adherence promotion. Figure 1 depicts the main intervention components and outcome measures as they relate to the PROSPR conceptual model.²⁹ Table 2 describes how the steps within each component differ between study arms.

Table 2.

Steps in osteoporosis screening and comparison who is responsible of study arms.

Step in Process	Usual Care	Bone Health Service (BHS)
Screening Promotion
Selection for screening	Discretion of provider	BHS RN queries EHR quarterly
Scheduling screening	Provider orders; clinic staff or Veteran phone call to radiology	BHS RN orders and coordinates scheduling with Veteran
Quantify risk from DXA results, determine if they meet treatment threshold	Discretion of provider	BHS RN and BHS MD based on FRAX risk
Shared decision-making with Veteran	Provider by phone or at next primary care visit	E-consult to provider, BHS phone call with Veteran, decision-tool sent to patient
Ordering treatment	Provider	BHS RN with provider co-signature
Adherence Promotion
Adherence monitoring	Discretion of provider	BHS RN telephone follow-up and personalized barrier assessment
Intervening when non-adherence detected	Discretion of provider	BHS RN uses adherence algorithm, educational visit

EHR = Electronic Health Record

Usual Care.

Primary care providers in usual care were given the VA Undersecretary Guidelines for primary osteoporosis screening and standard patient education materials for adherence support. Currently there are no metrics or tools for osteoporosis management in VA; therefore, this arm represents a usual practice control group.

Bone Health Service Model –

Patients in primary care teams randomized to the BHS model will have osteoporosis screening, education, and follow-up handled centrally by the bone health team. Providers can opt out of the service for patients in whom they believe it is not appropriate but are not responsible for most activities.

Screening Promotion

1. The Bone Health Nurse (BHS RN) identifies all patients eligible for screening using an EHR report run from the regional data warehouse. This report is programmed in SQL, and uses age, gender, ICD diagnosis codes, pharmacy records, and DXA orders to identify men eligible for primary osteoporosis screening. If more than 100 patients meet screening criteria, they are ordered by OST score³⁷ and the 100 highest risk are selected. Patients are contacted via letter and up to 2 telephone calls, and if they agree are scheduled for DXA.

2. The BHS RN obtains DXA results, calculates Fracture Risk Assessment Tool (FRAX) score for those with osteopenic T-scores³⁸, and patients who meet NOF treatment criteria are referred to the Bone Health MD for e-consult.

3. The Bone Health MD reviews additional clinical information in the EHR and generates an e-consult containing recommendations for additional laboratory evaluation (if needed) and treatment; this note is co-signed by the primary care provider. The BHS RN then contacts the patient for shared decision making and education, and places orders for provider co-signature.

Patient sample flow and outcome.

The sample is composed of all selected from initial screen and contacted for osteoporosis screening, projected to be 60–100 patients per team. For outcomes incorporating time, ‘start time’ will be set at the point of first contract irrespective of ultimate agreement to participate in the osteoporosis screen. To allow for patient flow through the team’s osteoporosis screen process, patients may be selected for study delaying first contact to provide immediate access to the osteoporosis screening slots. ‘End time’ will be defined as end of study or censoring due to death, date of nursing home/palliative care admission or date of last contact for those lost to follow-up.

Adherence Promotion.

All patients initiating oral bisphosphonates are called by the BHS RN at 1, 6, and 12 months to identify adherence barriers using a validated tool modified for osteoporosis therapy.²⁸ Algorithms for overcoming patient and health-care system barriers are used in these calls.^{32, 34, 39} For example, patients reporting gastrointestinal distress are offered annual intravenous therapy; patients with difficulty remembering a weekly medication are signed up for automated text message reminders. The 12-month call will ensure that the medication has been re-ordered. Subsequently, patients not refilling medications are identified via an EHR report that identifies gaps in refill requests, with telephone follow-up as indicated.

Staff Training and Supervision.

Study staff include a masters-level Project Director and a research RN who serves as the BHS RN. The BHS RN completed the National Osteoporosis Foundation Certification program, an online bundle of 17.25 credit hours covering 16 fracture prevention topics. The Project Director received additional training in qualitative analysis. A Research Assistant blinded to treatment assignment will complete outcomes assessment. The PI supervises staff weekly.

Intervention Fidelity.

Treatment fidelity is supported with Study Operating Procedures for each BHS element, and assessed using monitoring checklists at least twice annually.

Follow-up.

The intervention period will be 2 years. Staggered start times over 12 months accommodate DXA availability and staffing limitations.

Outcomes.

Outcome definitions, timing, and data sources are in Table 3.

Table 3.

Study outcome measures and the pre-specified clinically important difference.

Outcome	Definition/Measure	Data Source/Timing	Important Difference
Patient (Panel) Level (Aim 1)
Screening Rates	Proportion of eligible men screened in last 12 months	EHR data warehouse at baseline, year 1 and year 2	25% increase from 6% Usual Care
Medication Initiation	Proportion of screened men meeting treatment threshold who receive at least 1 prescription	EHR, intervention period year 1 and year 2, non-VA medication lists by chart abstraction.	30% increase from 55% Usual Care
Medication Implementation	Days of medication dispensed divided by follow-up days	Pharmacy dispensing records, for patients started within prior year at baseline, year 1 and 2. Non-VA medication lists by chart abstraction.	20% increase in MPR≥80%40 from 30% Usual Care
Medication Discontinuation	Time between first prescription dispensing date and the date of first medication possession gap of ≥3 months		20% difference
Harms	Proportion of men started on oral medication for new GI distress in 3 months Subtrochanteric fractures or Osteonecrosis of the jaw	ICD10 codes, new prescription for proton pump inhibitor or H2 blocker	15% increase from 30% Usual Care >expected 1/50,000 patient years treatment
Fractures (exploratory)	All clinical fractures excluding facial, digital	EHR, confirmed by chart abstraction	10% decrease from 2.5/100 person years Usual Care
Provider/Facility Level (Aim 2)
DXA volume	DXA orders/ 1000 patients/year, by intervention group	EHR, year 2
Bone Disease clinic volume	Consults/ 1000 patients/year, by intervention group	EHR, year 2
Primary care team satisfaction, time	Nominal Group Technique at Routine Staff meeting	Measured at 2 years
Health System/Policy Level (Aim 3)
Program Cost Effectiveness	Cost/quality adjusted life years (QALY) of the screening models compared to usual care	Markov model with above outcomes, fracture rates, cost and quality of life from VA and medical literature	Probability of cost/QALY >80% at thresholds of $50,000, $100,000, $200,00041

EHR = Electronic Health Record; MPR = Medication Possession Ratio

Patient-level.

Most patient-level outcomes will be extracted from EHR data. Outcomes not in coded fields (DXA results, non-VA medication prescriptions) will be obtained via chart abstraction by a research assistant masked to treatment assignment; a random 10% sample will be independently abstracted by a second team member to ensure >90% interrater agreement for key variables. Medication adherence will be assessed using ABC taxonomy categories of initiation, implementation, and discontinuation.⁴² While most patients fill prescriptions for osteoporosis medications with the VA pharmacy, non-VA prescriptions will be identified by chart abstraction; a study of VA primary care patients reported that <3% of prescription medications for chronic conditions were missed when both VA and non-VA medications were considered.⁴³ Fractures are an exploratory outcome due to limited power and incomplete fracture ascertainment as Veterans may seek care outside VA for fractures. However, under-ascertainment should be equal across the groups. As a surrogate outcome measure for fracture, bone mineral density in a random sub-sample of Veterans eligible for screening will be compared across treatment arms; the top 100 highest risk Veterans in each team by OST score will be contacted in random order until 25 per team have been recruited to receive a study DXA. Informed consent will be obtained from the Veterans for this measure.

Harms related to screening will be actively measured. Randomized trials of screening in women have found no detectable increase in anxiety related to DXA.⁴⁴ However, esophagitis is a known side effect, and gastrointestinal symptoms are commonly reported in oral bisphosphonate users although they occur at the same rate in those on placebo in randomized trials. We will evaluate the proportion of eligible Veterans with a new prescription for a proton pump inhibitor or H2 blocker during the study period. Rare side-effects including osteonecrosis of the jaw and atypical femoral fractures are not expected in this study as they usually occur after 5 or more years of treatment. For economic models, rare harms will be modeled using published population rates.

Provider/Facility Level.

Rates of DXA orders and bone disease clinic volume (where treatments other than oral bisphosphonates are administered) per 1000 patient years will be calculated from the EHR. Because surveys of primary care providers yield low response rates, we will use a Nominal Group Technique to obtain provider perspectives on the screening models.^{45, 46} The Nominal Group Technique combines quantitative and qualitative data collected from small groups of 8–20 people to rapidly generate a prioritized list of responses to specific questions during a 45–60 minute facilitated meeting. Questions will elicit providers’ perceived barriers and facilitators to osteoporosis care, experience with their assigned model, and estimated time spent per month on osteoporosis screening and management.

Health System/Policy Level.

Cost-effectiveness outcomes include the incremental cost per Quality Adjusted Life Year (QALY) for the BHS model relative to usual care generated from Markov simulation models as described below. These models require point estimates and sensitivity analysis ranges for transition probabilities, utilities and costs in order to generate reliable cost-effectiveness estimates. The proportion of eligible patients per panel, screening rates for each model, proportion meeting treatment criteria, treatment and adherence rates will be calculated from this study, with +/−1 SD used in sensitivity analyses. BHS model costs will be measured directly using staffing rates and primary care provider time estimates. Published ranges of screening costs, bisphosphonate effectiveness for primary prevention in men, and health utilities associated with major fracture types from systematic reviews will be used with their 95% confidence intervals.^47–50 Bisphosphonate average costs will be obtained from VA pharmacy data. Fracture rates from a prior study of over 4,000,000 men receiving primary care in VA will be used, with published rates from NHANES employed for sensitivity analyses.⁵¹

Covariates will be extracted from the EHR. Demographics include age, self-reported race, body mass index, and rural zip code as classified by Rural-Urban Commuting Area. Co-morbidities related to fracture risk include chronic lung disease, diabetes, endocrine disorder (hyperthyroidism, hyperparathyroidism, Cushing’s, hypogonadism), prostate cancer, rheumatoid arthritis, Parkinson’s disease, stroke, gastrectomy/malabsorption, smoking, alcohol abuse, chronic kidney disease, chronic liver disease, dementia. Medications include calcium, vitamin D, glucocorticoids, androgen deprivation therapy, traditional anti-epileptic drugs, proton pump inhibitors, selective serotonin reuptake inhibitors, and psychoactive medications.

Analysis

General Considerations:

As listed in Table 3, at a patient level the outcomes are a mixture of continuous (MPR) and binary (screening, medication initiation, discontinuation, harms, and fractures) outcomes. Several analytic issues must be considered. First, to follow the dictum, ‘analyze and you randomize’, the analytic structure must incorporate teams as the unit of analysis. The analysis will include the random effect, team, in the analysis to assess the fixed effect of interest, group. However, the effect of the clustering itself will be of interest, if, for example, characteristics of teams can predict risk or level of the outcome. Second, due the rolling enrollment patients will have varying ‘on-study’ times, possibly ranging between 2 and 3 years. Varying time-on-study times for these outcomes, require incorporation into the analytic structure as the probability of an event is impacted both by the risk and the time at risk. For the MPR, the outcome can be defined as average MPR/month, while for binary outcomes, the likelihood can be expressed using time-on-study as an offset, under, for example, Poisson regression. As noted above, use of Poisson regression, incorporating time at risk and team clusters are easily incorporated into the analytic structure. The analysis of each outcome listed below, will employ a 0.05 level alpha (two-tailed) criteria.

Primary outcomes:

(1). Osteoporosis Screening.

To incorporate the different time-on-study by patient, the group difference, BHS vs. UC, will be estimated using Poisson regression using GEE as listed above. The random effect of cluster will be included into the analytic structure. Using UC as the reference, the resulting Log Relative Risks (logRR) will be presented exponentiated relative risks (RRs) with 95% Confidence Intervals (CI) and p-values. To assess mediation/moderation, follow-up analyses will incorporate individual demographic and clinical variables and team characteristics, as listed above, into the model.

(2). Medication indication.

Similar to above, in the subset of men who are screened, the UC-BHS difference in the proportion for whom were dispensed and received an initial osteoporosis medication will be tested by Poisson regression, to allow for differing lengths of time per patient. The RRs, p-values and 95% CIs will be presented. In addition, follow-up analyses incorporating characteristics of the patient and team will assess moderating and mediating effects of these variables on the estimates.

(3). Medication Implementation.

We will employ the MPR scale as the outcome. The MPR measures the proportion of time covered by prescription, and allows analysis of differing lengths of time on study. We will assess the MPR in 3 periods, the first, recruitment-screening year, and the 2 follow-years. With 3 replicate correlated measures per person, Mixed Models will be employed. Relative standard repeated measures ANOVA, Mixed Models allows estimation in the presence of missing values, and incorporation of the random factor, team, presents no issues analytically. Initially, we will assess the correlation structure and incorporate the results into the analytic structure. Subsequently, we will test for a group by time interaction (on 2 df), and, if significant, we will compare the group effect at each timepoint, adjusting for 3 tests using Tukey adjustment. If the time by group interaction is non-significant, we will test the main effects of group and time. Estimates of group differences, 95% Cis, and p-values will be presented. The MPR may not be normally distributed, and may be zero-inflated. If such non-normal distributions are present, we will adapt the longitudinal models to assess, for example zero-inflated Poisson distribution, estimated under Generalized Estimating Equations (GEE). At the end of analysis, we will again assess moderation and moderation by characteristics of patient and team.

Secondary Outcomes:

As listed in Table 3, we show several outcomes measuring process and events. We are underpowered to test group differences by inferential techniques. Nonetheless, we will show and discuss group differences.

Medication discontinuation will be assessed by survival analysis (Proportional Hazards (PH)) to assess the time to first discontinuation of greater than 3 months, among those men initially dispensed drug. The model will incorporate clustering for team. The proportionality assumption which underlies the PH model will be assessed, and, if indicated a time by Group interaction will be included Harms will be assess among the men dispensed drug, the proportion listed for ICD codes for proton pump inhibitor or H2 blocker.

Fractures –

Similar to the modeling for medication discontinuation above, for the entire patient cohort, time to first fracture will be assessed by the PH model. Underpowered to declare significance, the effect of group, screening, drug initiation, demographic and clinical variables will be tested individually to assess a signal.

Aim 3.

To assess the economic impact of the BHS model, a Markov cost-utility model with 6 health states (Figure 3) will be constructed comparing different strategies of DXA screening, followed by 5 years of treatment with alendronate (the most commonly prescribed medication) for those with a femoral neck T-score of ≤−2.5 or FRAX 10-year fracture risk above current treatment thresholds (3% hip, 20% major osteoporotic fracture). For the base-case analyses, the model will be run for five different starting ages (65, 70, 75, 80, and 85) using Monte Carlo simulations with 40,000 trials each. Running the models for 5-year increments will allow us to adjust the transition probabilities for observed differences over time (e.g., higher fracture rates or different adherence rates with older age). The health states in the model will be no fracture, post-distal forearm fracture, post-clinical vertebral fracture, post-hip fracture, post-other fractures (humerus, scapula, ribs, pelvis, distal femur, pelvis, patella, tibia, or proximal fibula), and death. Following the recommendations of the 2^nd Panel on Cost-Effectiveness in Health and Medicine,⁵² we will analyze our model from both the VA and societal perspectives and present our results as an impact inventory.

Figure 3.

Markov model for assessing model cost-effectiveness

The direct and indirect costs of fractures will be assigned as a transition cost. The disutility associated with these fractures will be modeled as a QALY decrement associated with that fracture state based on a published systematic review of fracture disutility rates⁵³ and is assigned for 6 months upon which they return to baseline state, except for hip and vertebral fractures for which permanent disutility is the norm. If an individual suffers an additional fracture, the costs of that fracture will be assigned, but the individual will remain in the post-vertebral or post-hip fracture state because the long-term disutility associated with prior vertebral and hip fractures is greater. Long-term care costs beyond the first year after hip fracture will be assigned as annual costs. The cycle length for our model will be 3 months and we will assume a maximum of 2 fracture types. A discount rate of 3% will be assumed for costs and health benefits.

Treeage software will be used for analyses. One-way sensitivity analyses will be performed varying discount rates, fracture rates, fracture costs, fracture disutility, costs of DXA, the onset and offset of fracture reduction benefit following initiation and cessation of drug therapy, medication adherence, the relative risks of fractures attributable to osteoporosis or prior clinical fracture, cost of screening and yearly bisphosphonate. Because of uncertainty regarding the nonvertebral fracture reduction efficacy of oral bisphosphonates for men, 2-way sensitivity analyses will be performed assuming reduced fracture efficacy. Probabilistic sensitivity analyses will be performed using 2^nd order Monte Carlo simulations in which each parameter value will be drawn from a distribution with characteristics specific to that parameter. For example, log-normal distributions of fracture direct costs and normal distributions of fracture rates and long-term care costs following hip fracture. The distributions of the relative risks of incident fractures associated with osteoporosis, prior fracture, and oral bisphosphonate therapy are assumed to be log-normal. Uniform distributions will be used to model variability in fracture disutility and indirect fracture costs. Results from the PSA will be presented as scatterplots on the cost-effectiveness plane and as cost effectiveness acceptability curves. We prespecify a willingness to pay threshold of $100,000 per QALY.

Power.

We estimate power using patient level outcomes. Because the study will operate under a waiver of informed consent, we do not account for refusal to participate; however, the expected Veteran refusal of screening and treatment rates are reflected in our assumptions. We expect a low Intraclass

Correlation Coefficient (ICC) for patients within primary care teams as patients are randomly assigned to providers, and based on prior studies demonstrating ICC<0.1.⁵⁴ Finally, unless otherwise noted, we determine power for the single degree of freedom contrast of import – Usual Care vs. BHS group across the follow-up time points. A sample size of 38 primary care teams provides adequate power to detect a clinically meaningful difference for all our primary patient-level outcomes, except and Harms fracture rates which are exploratory in this analysis. We recruited 38 teams but expect attrition of as many as 3 teams over 2 years.

(1). Screening.

We assume a 6% osteoporosis screening rate in usual care based on preliminary data. Ignoring the clustering by teams, assuming 3800 patients, setting the Type-I error rate (alpha) to 0.05 (two-tailed), as seen in table 3, we have greater than 99.9% power if we observe a 25% increase (6+25=31%) relative to UC. Indeed, relative to the 6% base, our design can detect increases of as small as 2.5 and 2.8% at 80 and 90% power respectively.

(2). Medication Initiation.

Among the screened (assuming 10% screening across the 2 groups, 0.1*3800=380) we assume 55% will receive medication. Setting power to 80% and a Type-I error rate of 0.05 (two-tailed), relative the expected 55% level, our design can detect and increase of 14 and 16% respectively at 80 and 90% power respectively. These values fall well below the clinical important 30% increase.

(3). Femoral Neck BMD.

We pre-specify that a clinically significant difference in BMD between Usual Care and BHS is 3% at the femoral neck and 5% lumbar spine difference. These are the differences observed between treatment and control in the pivotal trial for approval of bisphosphonate treatment in men, which demonstrated a reduction in vertebral fractures.⁵⁵ We assume average femoral neck BMD men 0.939g/cm² (SD 0.133)⁵⁶. To detect a between group difference of 3% (0.967), UC vs BHS, we will require a total N of 712 (356 each in UC and BHS), at alpha=0.05 (two-tailed) and a power of 0.80. To account for scans that are not interpretable due to hardware, osteophytes, quality problems etc. we will recruit n=900 to ensure an adequate analytic sample. We also plan to perform a secondary-sensitivity analysis of the impact of treatment modality on lumbar spine BMD. This outcome will be modeled equivalently to the femoral neck analysis, with a Type-I error rate set to 0.05 (two-tailed).

Missing data Strategies.

For continuous measures where zero is illogical, we will impute a value, simulate and use bootstrapping to derive estimates of effect with appropriate standard errors. To implement the multiple imputation techniques for primary analyses, we will use preliminary analyses of such associations, including graphical displays, to investigate the plausibility of the assumptions underlying the imputation model employed.⁵⁷ Further, we will use 25 or more imputed data sets, with FU bootstrapped estimates in order to reduce the impact of the random sampling.

Discussion

An efficient and effective care model that improves osteoporosis screening and adherence is urgently needed to address the national “crisis in osteoporosis treatment”.⁵⁸ Presently, there are no available randomized trials of osteoporosis screening in men, and very few studies of the impact of a centralized BHS model for primary fracture prevention in either gender^59,60 Because of this lack of evidence, the United States Preventive Services Task Force and the VA National Center for Health Promotion and Disease Prevention have concluded that there is insufficient evidence to make a recommendation for or against screening in men. Nevertheless, a VA national cohort study showed screening was associated with significantly lower fracture rates in high-risk subgroups, economic models based on cohort studies suggest that screening is likely to be highly cost-effective for both genders,^61–64 and professional groups including the VA Undersecretary for Health advocate for primary osteoporosis screening in high-risk men.^65,38 This clinical trial will therefore provide important information on the impact of a centralized screening model (BHS) on bone mineral density, healthcare utilization and cost that be vital for policy makers and clinicians.

The BHS model represents a significant departure from current practice and would require greater resources than a more traditional practice management model. We originally designed a 3-arm study that would have allowed us to estimate the impact of both models compared to usual care and each other. Although the COVID-19 pandemic necessitated the early cessation of the practice management arm, we believe that even without such disruption a BHS model is likely to be the preferred approach for primary osteoporosis screening and treatment. Practice management models are more difficult to initiate and sustain across large health systems due to high staff turnover and shifting team priorities. Indeed, even before the pandemic we experienced a lower-than-expected primary care team recruitment rate, with the most frequent reason cited for declining being the additional time required for teams in the practice management arm. In contrast, the BHS model reduces the burden on primary teams by taking on much of the shared decision making, patient education and monitoring. This study will quantify the impact on patients, providers, health systems, and costs such that health system leaders will be able to make informed decisions on adoption.

There are a number of threats potentially impacting this protocol. Patients occasionally transfer between primary care teams, potentially leading to contamination. We will use an intention to treat analysis strategy such that patients are analyzed in the group to which their current team is assigned, but monitor team changes for a per protocol analysis if contamination rates are high. By enrolling in the study and receiving current guidelines, providers in the usual care group may modify their practice. However, multiple prior studies within and outside of VA have shown little change during osteoporosis quality improvement programs,^{16, 17} and the pre-post design allows us to measure the impact of study observation on provider behavior. Some Veterans have non-VA providers who may order DXA through Medicare. However, Medicare reimburses for primary osteoporosis screening in men for a restricted number of secondary diagnoses; in our VA national study >80% of DXA screening was completed through VA, rather than CMS. Outside DXA results scanned into the EHR will be identified through chart abstraction.

The study is expected to inform national Osteoporosis Screening Guidelines for men by quantifying differences in bone mineral density among those at highest risk between screening arms. BMD is a validated surrogate endpoint for fracture that is under consideration by the FDA for approval of new osteoporosis pharmacotherapies. Thus, this study is positioned to be one of the first randomized trials powered to assess the impact of primary osteoporosis screening in men.