Introduction

Despite the availability of published clinical practice guidelines and evidence of the efficacy of a healthy lifestyle and pharmacotherapy in osteoporosis prevention and treatment [1–6], studies consistently show the disease is underdetected and undertreated, even in high-risk patients [7–9].

In the effort to overcome the care gap in the detection and management of osteoporosis, many studies have used controlled study designs to evaluate interventions aimed at improving osteoporosis detection and treatment [10–26]. Results have been mixed: some studies found relatively large intervention effects [10, 14, 15, 19], some observed modest outcomes [11–13, 17, 18, 20–23], while others found no effect [16, 24–26]. A systematic review focussing on the effectiveness of these interventions may therefore provide some insights into which are the most effective strategies to adopt.

A systematic review conducted in 2008 concluded that multi-component tools targeting both physicians and patients may be effective in supporting clinical decision making in osteoporosis disease management [27]. However, the search performed for the review was limited to randomized controlled trials (RCTs) published from 1966 to 2006, and it thus excluded a substantial number of studies published from 2007 through 2009. Another recent systematic review, which was conducted by Lai et al. [28], observed that osteoporosis interventions by healthcare professionals were associated with improved quality of life, medication compliance, and calcium intake. However, the review found no effect in terms of changes in bone mineral density (BMD), medication persistence, knowledge, or other lifestyle modifications. These reviews assessed the effectiveness of the interventions for patients with established osteoporosis, though—not for people at risk for whom screening is indicated, a population for which osteoporosis interventions might best be put to use to prevent or delay the onset and consequences of the disease. An updated systematic review of interventions targeting patients at risk or at high risk for osteoporosis and fractures and candidates for osteoporosis screening or treatment thus seems to be very much in order.

The objective of this study was to assess the effectiveness of interventions aiming at improving the detection and treatment of osteoporosis in primary care regarding the incidence of BMD testing, osteoporosis treatment initiation and fractures in patients at risk and at high risk in whom osteoporosis screening or treatment is indicated under Canadian and American guidelines [1–6]. To do so, we first systematically reviewed the literature in order to identify studies that evaluate interventions designed specifically to improve osteoporosis detection and treatment in primary care and, second, we quantitatively combined their results, when applicable. More specifically, patients of interest in this review include patients in whom osteoporosis screening is indicated because of the presence of risk factors for osteoporosis and patients in whom osteoporosis treatment is recommended due to a previous fragility fracture or chronic glucocorticoid use. We expected higher incidences of BMD testing and osteoporosis treatment initiation and a lower incidence of fractures following interventions designed to improve osteoporosis detection and treatment.

Methods

Data sources

The study was based on an a priori protocol (available online at http://www.recherchepl.ca/autres-productions-scientifiques.php) which prespecified research objectives, search strategy, study eligibility criteria, and methods of data extraction and statistical analysis. The findings are reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement [29].

Studies were identified by searching MEDLINE (1950–2009), EMBASE (1980–2009), PsycINFO (1967–2009), ERIC (1965–2009), All EBM Reviews (which includes the Cochrane Database of Systematic Reviews, ACP Journal Club, the Database of Abstracts of Reviews of Effects, the Health Technology Assessment database, the National Health Service Economic Evaluation database, and the Cochrane Methodology Register database), CENTRAL (1991–2009), CINAHL (1981–2009), and Current Contents (1993–2009). We also searched the gray literature for unpublished and “in progress” studies on websites of clinical trial registries (the CenterWatch Clinical Trials Listing Service, the Current Controlled Trials International Standard Randomised Controlled Trial Number Register and metaRegister of Controlled Trials, and the World Health Organization International Clinical Trials Registry Platform Search Portal), the Turning Research Into Practice database, Digital Dissertations (ProQuest), the Canadian Institutes of Health Research website, the National Institutes of Health Research Portfolio Online Reporting Tool, and the proceedings of the International Osteoporosis Foundation World Conference on Osteoporosis from 2000 to 2008. The reference lists of the articles selected and relevant reviews [27, 30–32] were also screened for further potentially eligible studies, and the corresponding authors of the studies included were contacted.

Search strategy

Electronic databases were searched using strategies incorporating selected MeSH terms and free-text terms combined with the methodological component of the Cochrane Effective Practice and Organisation of Care (EPOC) group search strategy (available at http://epoc.cochrane.org/specialised-register). The MEDLINE search strategy (see Appendix A) was adapted for the other databases. All search strategies were carried out with the assistance of an experienced librarian and peer reviewed by a second experienced librarian.

Study selection

Studies were included if they were RCTs (patient-randomized trials and cluster-randomized trials), controlled clinical trials or quasi-randomized trials, controlled before-after studies, or interrupted time series studies evaluating interventions aimed at improving the detection and treatment of osteoporosis. Quasi-randomized trials were defined as studies in which participants are assigned prospectively to study groups using a quasi-random allocation method, such as alternation or date of birth. Published and unpublished studies written in English or French were considered. We restricted our search to the years 1985–2009 because, since the first guidelines regarding the use of hormone replacement therapy (HRT) to prevent osteoporosis in post-menopausal women were published in 1985 [33], it is unlikely that quality improvement interventions in osteoporosis were conducted before that date. The duration of patient follow-up was limited to at least 3 months after the intervention, which corresponds to the minimum period needed to capture patients starting osteoporosis medications [18]. Abstracts of eligible studies for which no full-text report is available could also be included in the review, but, given the sparseness of data, these publications could not be considered in the principal data analysis.

Interventions of interest included those specifically designed to improve the detection and treatment of osteoporosis, such as educational lectures or meetings, training workshops, educational outreach visits, written educational material, audit and feedback, computerized decision aid support tools (such as electronic prompts or reminders), lists of at-risk patients, patient-risk assessment, and patient-mediated interventions. Furthermore, eligible interventions had to take place in a primary care setting and involve or target primary care physicians, patients, primary care nurses, community pharmacists, or a combination of these populations. In addition, eligible studies had to compare the intervention group to a comparison group receiving either the usual care or a control intervention on a topic other than osteoporosis (e.g., an educational lecture on cholesterol management). A comparison group receiving printed material on osteoporosis was considered a “usual care” group. Interventions that were a component of a general intervention for chronic diseases were excluded, as were studies evaluating the efficacy of specific medications for osteoporosis (e.g., bisphosphonates), studies on exercise and physical activity programs and studies assessing interventions to improve adherence to osteoporosis treatment. Fall-prevention interventions were also excluded, unless they included a component aimed specifically at the detection and treatment of osteoporosis. Interventions geared to specialists (e.g., rheumatologists, orthopedic surgeons) and inpatient interventions were excluded, unless a component involved primary care physicians. Interventions targeting nursing home patients were not included.

To be eligible for inclusion in the review, studies had to evaluate the intervention in a population of patients either (1) at risk for osteoporosis and fractures, including women aged 65 years or older, men aged 70 years or older, and men/women aged 50 years or older with at least one major risk factor for osteoporosis (family history of osteoporosis, malabsorption syndrome, primary hyperparathyroidism, hypogonadism, or early menopause), with no previous BMD testing and no current osteoporosis treatment; or (2) at high risk for osteoporosis and fractures, including men/women of any age receiving a daily dose of at least 5 mg prednisone or equivalent for more than 3 months or with a previous fragility fracture with no current osteoporosis treatment. This population corresponds to patients for whom osteoporosis screening or treatment is indicated under published Canadian and American guidelines [1–6].

Outcomes of interest included the incidence of BMD testing; osteoporosis treatment initiation with a bisphosphonate, raloxifen, calcitonin, teriparatide, or HRT; initiation of calcium/vitamin D supplements; BMD testing and/or osteoporosis treatment initiation (composite endpoint); and fractures. These outcomes were measured at the longest follow-up time point reported.

Two investigators (MCL and GJ) independently reviewed the titles and abstracts of the studies found. The inclusion and exclusion criteria were then applied independently by six evaluators using a standardized eligibility evaluation form; each potentially relevant article was assessed by two evaluators. Eligibility criteria were assessed in the following order: type of intervention, study design, type of participants, duration of patient follow-up, and type of comparator. The first “no” response served as the primary reason for exclusion. Assessors were not blind to any information, and disagreements were resolved by discussion to reach consensus. Authors of studies for which the report was incomplete (i.e., for which information was missing) were contacted.

Data extraction and risk-of-bias assessment

Data extraction was performed by two independent investigators (MCL and GJ) using a standardized data extraction form. The following characteristics and data were extracted from each study selected: study design, unit of allocation, allocation sequence generation method, allocation sequence concealment method, country, study population, participant inclusion and exclusion criteria, sex of included patients, description of interventions, format of interventions, duration of interventions, types of participants involved in interventions, description of the control group, duration of follow-up, data collection methods, unit of analysis, intent-to-treat analysis, total number of patients and health professionals entered in the study, randomized (if applicable), lost to follow-up (with reasons) and included in analyses in each study group, number and proportion of patients experiencing each outcome of interest in each study group, crude and adjusted risk ratios and odds ratios with confidence intervals (CIs) and/or p value reported for each comparison regarding outcomes of interest, reported intra-class correlation coefficients (ICC; for cluster RCTs), and source of funding. Also documented were any elements of the Chronic Care Model [34] in the intervention. The Chronic Care Model was conceptualized by Wagner. It is a multi-dimensional guide to developing effective chronic care. It centers on six key elements of care, including: (1) self-management support (e.g., self-help or peer support groups, patient education classes, self-management tools such as flowcharts on which patients record their own laboratory results, self-management plans with self-improvement goals); (2) decision support (e.g., reminders based on evidence-based guidelines, continuing medical education, academic detailing); (3) clinical information systems (e.g., creation of computerized registries to increase data accessibility, data sharing between health professionals, data sharing with patients); (4) healthcare organization (e.g., inter-professional collaboration, visit planning for continuous follow-up); (5) community resources and policies (e.g., efficient community programs, increasing collaboration between health professionals and community resources, improving the use of non-medical resources); and (6) delivery system design (e.g., creating practice teams with a clear division of labor, planned visits, training non-physician personnel to support patient self-management, arranging for routine periodic tasks). Disagreements were resolved by discussion or, if they persisted, by arbitration with a third assessor (LL or SP). Data from multiple reports of the same study were extracted on the same data extraction form.

The data extraction form incorporated a risk-of-bias assessment form adapted from the Cochrane Collaboration’s risk-of-bias assessment tool [35] and the Cochrane EPOC group quality checklist (available at http://epoc.cochrane.org/epoc-resources-review-authors). The risk of bias was assessed on the basis of the percentage of applicable criteria that were met and addressed allocation sequence generation, allocation sequence concealment, completeness of follow-up, blinding, baseline outcome measurements (health professional performance or patient outcomes prior to the intervention), reliability of outcome measures, possibility of contamination, possibility of unit-of-analysis error (for cluster RCTs), participants’ baseline characteristics, incomplete outcome data (likelihood that missing outcome measures could bias the results), and possibility of other risks of bias. The risk-of-bias assessment was conducted at the same time and in the same way as the data extraction.

Data synthesis and analysis

Analyses were stratified according to the type of patients involved in the studies (patients at high risk for osteoporosis and fractures, namely men/women using oral glucocorticoids for ≥3 months or with a previous fragility fracture; at-risk patients, including women aged 65 years or older, men aged 70 years or older, and men/women aged 50 years or older with at least one major risk factor for osteoporosis; or both high-risk and at-risk patients) and were performed by intention to treat. If a study had more than one intervention group, we selected the one with the most intensive intervention (i.e., with the greatest number of components) in order to ensure the independence of the pooled calculations.

For each study included, the risk difference (RD) between intervention and control groups regarding each outcome of interest and the corresponding 95% CIs were computed. In studies with a cluster RCT design, adjustment was made for the design effect: the proportion of patients experiencing each outcome of interest during follow-up in each study group was divided by the design effect, which may be calculated by the formula 1 + (M-1) ICC, where M is the average size of each cluster and ICC is the intra-class correlation coefficient [36]. If the ICC was not reported in a paper, the authors were contacted. If the ICC remained unknown, an ICC of 0.01 was assigned, and sensitivity analyses were performed to explore the impact of different values (0.05, 0.10, and 0.15). Our assumptions are reasonable, given that following an osteoporosis workshop for family physicians in a previous cluster cohort study, ICCs of 0.01 and 0.03 were observed for the incidence of BMD testing and the incidence of osteoporosis treatment initiation, respectively, in patients at-risk for osteoporosis and fractures [23].

Sources of heterogeneity were explored by examining the impact of the risk of bias in the study (studies having a proportion of applicable risk-of-bias criteria fulfilled below vs. over the median proportion of applicable criteria fulfilled in all the studies included) and study design (e.g., patient RCTs vs. cluster RCTs) on results in sub-analyses. Heterogeneity within studies targeting similar types of patients (high-risk patients vs. at-risk patients) was assessed with the chi-square (Q statistic) and quantified using the I 2 statistic, which indicates the proportion of variability across studies due to heterogeneity rather than sampling error. In groups of studies without substantial heterogeneity (with a chi-square statistic having a p value > 0.10 and an I 2 statistic less than 50%), pooled RDs were calculated using a Mantel–Haenszel random effects model. Statistical analyses were performed using Cochrane Review Manager 5 (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2008).

Results

Our search identified a total of 9,545 citations. They were screened for relevance, and 91 were selected for full-text review (Fig. 1). In all, 16 reports regarding 13 different studies met the inclusion/exclusion criteria and were included in the systematic review [10–18, 37–43]. Studies were excluded for not meeting the criteria for target intervention (five studies), study design (19 studies), patients (37 studies), follow-up duration (one study), and comparator (eight studies; see Appendix B for the complete list of excluded studies with primary reasons of exclusion). A total of 19 studies (12 published and seven ongoing) were excluded because they possibly included patients who already had undergone BMD testing or were already receiving osteoporosis treatment; more specifically, five studies included patients who already had undergone BMD testing [21, 44–47], one study included patients who were already receiving osteoporosis therapy [48], seven studies included patients who already had undergone BMD testing, were already receiving osteoporosis treatment or had an osteoporosis diagnosis [25, 49–54], and six studies did not report whether patients who already had undergone BMD testing or were already receiving osteoporosis treatment were excluded or not [20, 24, 55–58]. Eligibility could not be assessed in five studies because of information missing in the reports we retrieved: for one study, the author did not respond to our messages; three studies lacked contact information for the author; and one study, according to the investigator contacted, was interrupted because of problematic patient recruitment. Only one screened study was excluded because of ineligible language (Chinese); based on the information in the abstract, the study did not meet the inclusion criteria for target study design and patients.

Fig. 1
figure 1

Study flowchart

Full size image

Study characteristics

The majority of studies selected involved patients at high risk for osteoporosis and fractures (8/13 studies), and most evaluated interventions targeting primary care physicians and their at-risk patients (12/13 studies; Table 1). Three studies used a quasi-randomized design [10, 14, 41], six were patient-randomized trials [12, 15, 18, 37, 38, 42], and four were cluster-randomized trials [11, 13, 16, 17]. Two studies were ongoing [38, 41], and two others were methodological papers describing the study protocol [37, 42].

Table 1 Characteristics of studies included in the review
Full size table

Interventions included educational material for patients (eight studies) [11–14, 16, 18, 37, 41], physician notification of patients’ osteoporosis and fracture risk (eight studies) [10, 11, 14, 15, 18, 37, 38, 41], patient notification of osteoporosis and fracture risk (six studies) [10–13, 17, 41], patient counseling (four studies) [14, 15, 18, 41], electronic medical record (EMR) prompts for physicians (two studies) [12, 13], list of at-risk patients provided to physicians (two studies) [13, 17], physician education by academic detailing (two studies) [16, 17], arrangements for BMD testing and bisphosphonate prescription by the study physician (one study) [15], a risk assessment tool provided to physicians and discussion of the tool with the study coordinator before appointments with at-risk patients (one study) [42], and peripheral BMD testing by quantitative ultrasound performed by the patient’s community pharmacist (one study) [18] (Table 1).

Study population also varied across studies: eight studies included both men and women [10, 14–18, 38, 41], and five involved women only [11–13, 37, 42] (Table 1). Most studies included patients with a previous fragility fracture (11 studies) [10–12, 14–18, 37, 38, 41], while others involved patients aged 65 years or older (four studies) [13, 16–18], oral glucocorticoid users (three studies) [16–18], patients with low BMD (one study) [42], a family history of osteoporosis (one study) [18], or early menopause (one study) [18]. Most studies targeted both the primary care physicians and their patients (10 studies) [11–17, 37, 38, 41]. One study involved patients, primary care physicians and community pharmacists [18], and another targeted patients, primary care physicians and orthopedic surgeons [10]. Only one study focused solely on primary care physicians [42].

In terms of the elements of the Chronic Care Model, most studies involved self-management support (10 studies) [10, 11, 13–16, 18, 37, 38, 41] and decision support (10 studies) [11–14, 16–18, 38, 41, 42] (Table 1). Other studies were related to clinical information systems (three studies) [12, 13, 17], delivery system design, and self-management support (one study) [15]. One study involved four domains: self-management support, decision support, clinical information systems, and healthcare organization [13]. Of the four cluster RCTs included in the review, none reported the value of ICCs; three authors provided these data, but one could not be contacted [11]. ICC values provided by the authors for outcomes of interest ranged from 0.02 to 0.04 [13, 16, 17].

Risk-of-bias assessment

Reported study quality was generally good; the median proportion of applicable criteria fulfilled was 73% (range from 30% to 90%), excluding ongoing studies [38, 41] for which much information was missing, and protocol description studies with no follow-up of patients [37, 42] (Tables 2, 3 and 4). None of the studies included in the review compared health professional performance or patient outcomes prior to the intervention (baseline outcome measurements).

Table 2 Risk of bias in studies included in the review—randomization, blinding, and possibility of unit-of-analysis error
Full size table
Table 3 Risk of bias in studies included in the review—completeness of follow-up, comparability of study groups, reliability of outcome measures, and other risks of bias
Full size table
Table 4 Overall risk of bias in studies included in the review
Full size table

Outcomes of interest

Eight studies provided data on outcomes of interest [11–18] (Table 5). Follow-up durations ranged from 4 to 16 months. In seven out of eight studies, the incidence of BMD testing was higher in the intervention group than in the control group; absolute differences ranged from 22% to 51% in studies targeting high-risk patients [11, 12, 14, 15], equalled 18% in the one study involving at-risk patients [13], and varied from 4% to 12% in studies targeting both at-risk and high-risk patients [17, 18]. In five studies, the incidence of osteoporosis treatment initiation was higher in the intervention group than in the usual-care group; absolute differences ranged from 18% to 29% in studies targeting high-risk patients [11, 14, 15] and equalled 4% in the one study involving at-risk patients [13] and 2% in a study targeting both at-risk and high-risk patients [17]. Four studies showed a statistically significant higher incidence of the composite endpoint (undergoing BMD testing and/or initiating osteoporosis treatment following the intervention) as compared to the control group; absolute differences ranged from 37% to 41% in studies targeting high-risk patients [12, 15] and from 4% to 11% in studies targeting at-risk and high-risk patients [17, 18]. Two studies reported statistically non-significant differences for initiation of calcium/vitamin D supplements [11] and fracture incidence [15].

Table 5 Results of studies included in the review regarding outcomes of interest
Full size table

The study showing the highest incidence of osteoporosis-related preventive practices was very intensive, and included arrangements for outpatient BMD tests and a bisphosphonate prescription written by a study physician and dispensed by the local community pharmacy [15]. The observed absolute between-group difference was 51% for BMD testing (80% in the intervention group vs. 29% in the control group; 95% CI: 39–61%), 29% for osteoporosis treatment initiation (51% vs. 22%; 95% CI: 17–41%) and 41% for the composite endpoint of undergoing BMD testing and/or initiating osteoporosis treatment (67% vs. 26%; 95% CI: 28–53%).

Stratification for the sex of study patients in the studies included in the meta-analysis revealed that interventions targeting both men and women produced results similar to those of interventions targeting women only. In studies with both men and women, absolute between-group differences ranged from 4% to 51% for BMD testing, from 3% to 29% for osteoporosis treatment initiation, and from 4% to 41% for the composite endpoint of undergoing BMD testing and/or initiating osteoporosis treatment [14, 15, 17, 18]. In studies with women only, absolute differences ranged from 18% to 28% for BMD testing and equalled 18% for osteoporosis treatment initiation and 37% for the composite endpoint [11–13].

In terms of the number of components in the intervention, one study of a two-component intervention reported no statistically significant differences in outcomes of interest [16]. In interventions with three components, absolute differences ranged from 4% to 45% for BMD testing [11, 12, 14, 17], 3% to 29% for treatment initiation [11, 14, 17], and 4% to 37% for the composite endpoint [12, 17]. Similar results emerged for interventions with four components: absolute differences ranged from 12% to 51% for BMD testing [13, 15, 18], equalled 29% for treatment initiation [15], and ranged from 11% to 41% for the composite endpoint [15, 18].

Meta-analysis

Substantial observed heterogeneity precluded a meta-analysis of the incidence of BMD testing following the intervention (Table 6). However, a meta-analysis was possible for the incidence of osteoporosis treatment initiation, which showed a significant 20% increase following intervention for high-risk patients (95% CI: 7–33%; Fig. 2a). Results were unchanged when the very intensive study by Majumdar et al. [15] was excluded (Fig. 2b). Moreover, data could be pooled regarding the incidence of the composite endpoint of undergoing BMD testing and/or initiating osteoporosis treatment, which showed a significant 40% increase following intervention in high-risk patients (95% CI: 32–48%; Fig. 3).

Table 6 Results of heterogeneity tests in primary and sub-analyses
Full size table
Fig. 2
figure 2

Effect of interventions aimed at improving detection and treatment of osteoporosis in primary care on incidence of osteoporosis treatment initiation in high-risk patients including (a) and excluding (b) the very intensive study conducted by Majumdar et al. [15]

Full size image
Fig. 3
figure 3

Effect of interventions aimed at improving detection and treatment of osteoporosis in primary care on incidence of BMD testing and/or osteoporosis-treatment initiation in high-risk patients

Full size image

Sub-analyses

Sub-analyses to explore heterogeneity in relation to the risk of bias in the studies included in the meta-analysis (studies with a low risk of bias fulfilling at least 73% of their applicable criteria, vs. studies with a higher risk of bias fulfilling less than 73% of their applicable criteria; 73% corresponds to the median proportion of applicable criteria fulfilled in the studies included in our review) and regarding study designs did not change the results (Table 6). Sensitivity analyses performed for the missing ICC values in one study [11] did not change the results (Table 6).

Discussion

Our systematic review included 13 studies evaluating primary care interventions designed to improve the detection and treatment of osteoporosis in patients at risk or at high risk for osteoporosis and fractures in whom osteoporosis screening or treatment is indicated. Most of the interventions included in our study were multifaceted, targeted both primary care physicians and their at-risk patients and often included educational material for patients, physician notification, and/or continuing medical education for physicians. Absolute differences in the incidence of BMD testing ranged from 22% to 51% in studies with high-risk patients and from 4% to 18% in studies involving at-risk and high-risk patients. Absolute differences in the incidence of osteoporosis treatment initiation ranged from 18% to 29% in studies with high-risk patients only and from 2% to 4% in studies involving at-risk and high-risk patients. Two studies reported statistically non-significant differences for initiation of calcium/vitamin D supplements and fracture incidence. Interventions with at least three components had a greater impact than interventions with two components on outcomes of interest. A meta-analysis pooling the results of four trials showed that the incidence of osteoporosis treatment initiation in high-risk patients increased by 20% following intervention (95% CI: 7–33%), and the combination of the data from two other studies showed that the incidence of BMD testing and/or initiating osteoporosis treatment in high-risk patients also increased by 40% with the intervention (95% CI: 32–48%). It has been observed that very costly and labor intense interventions are generally efficacious, but these interventions are not generalizable to most clinical settings.

A previous systematic review by Kastner and Straus [27] centering on the effectiveness of clinical decision support tools for osteoporosis disease management was published in 2008. Study heterogeneity prevented the authors from statistically combining the results of individual studies. Still, they concluded that multi-component tools targeting both physicians and patients may be effective in supporting clinical decision making in osteoporosis disease management. The review was limited to RCTs published from 1966 to 2006, though, and so excluded many studies on osteoporosis interventions published from 2007 through 2009. (The present review included 10 studies published or conducted between 2007 and 2009). Moreover, the 2008 review assessed the effectiveness of osteoporosis management tools in men and women with established osteoporosis (i.e., with a confirmatory diagnosis of osteoporosis or an existing or previous fragility fracture), as opposed to patients at risk for osteoporosis for whom screening is indicated. Lai et al. [28] conducted a systematic review of osteoporosis interventions targeting community-dwelling postmenopausal women with osteoporosis. They observed that the interventions were associated with improved quality of life, medication compliance, and calcium intake, although no effect was observed in terms of changes in BMD, medication persistence, knowledge, and other lifestyle modifications. Again, their review targeted only women with established osteoporosis. Furthermore, it did not assess the effect of interventions on osteoporosis screening and treatment.

The small number of studies precluded the use of meta-regression models. Therefore, the differential effectiveness of the various components of the interventions could not be investigated. However, previous systematic reviews suggest that passive dissemination of information such as traditional continuing medical education seems to be generally ineffective, while the use of computerized decision support systems, educational outreach visits, audit and feedback, and patient-mediated interventions seem to improve general professional performance at a greater extent [59–64]. Furthermore, the results of the present systematic review and meta-analysis are in line with these previous reviews regarding interventions aiming at improving general professional performance; they suggest that multifaceted interventions tend to be more effective than single interventions but often lead to only modest improvements in professional practice [59–64].

In the specific area of osteoporosis, numerous barriers to the application of guidelines by primary care physicians have been identified at the patient, provider, and healthcare system levels [65]. Barriers at the patient level include denial of osteoporosis diagnosis and risk factors, lack of awareness of osteoporosis treatment and prevention therapies and their efficacy, and lack of understanding of the potential morbidity and mortality of untreated osteoporosis [65]. Barriers at the physician level include lack of recognition of fragility fracture events as osteoporosis-defining events; low prioritization of osteoporosis for patients with multiple comorbidities; resistance to change; lack of awareness of the morbidity, mortality, and healthcare costs associated with osteoporosis; uncertainty about the indications for and interpretation of BMD testing; reluctance to start a new treatment in seniors already taking many medications; lack of time; and competing demands during appointments [65, 66]. Finally, barriers at the healthcare system level include the static nature of traditional healthcare processes; lack of system-wide standard orders; insufficient coordination of care between subspecialty and primary care providers; unwillingness of physicians to assume responsibility for preventive care; and fragmented financing for preventive care [65].

In light of these findings, it can be hypothesized that the involvement of various health professionals such as community pharmacists, physician assistants, and nurse practitioners in addition to primary care physicians in future interventions might help address barriers at the patient level by educating patients about the gravity of osteoporosis, its risk factors, and its therapies. Broader professional involvement might also help in dealing with barriers at the physician level by providing physicians with guideline-based recommendations on osteoporosis screening and treatment following the assessment of patient risk factors. A recent RCT included in this review notably assessed a community–pharmacist-driven intervention in which a pharmacist evaluated patient risk factors, made recommendations regarding BMD testing and pharmacotherapy, and forwarded recommendations to the primary care physician [18]; however, the study yielded only modest improvements. Targeting community pharmacists more intensively and giving them a greater role in osteoporosis detection and management may be an interesting avenue for developing future strategies.

To reduce the probability of bias, the rigorous methodology of our systematic review restricted the selection of studies to designs with the highest internal validity (e.g., RCTs). We conducted a comprehensive search of the literature using well-defined terms and inclusion/exclusion criteria, and we had articles evaluated at each level of selection by two independent assessors using standardized forms. We also addressed potential sources of variability between studies by assessing the risk of bias and by exploring differences in the populations of patients targeted, interventions, populations involved in interventions, study designs, and outcomes.

This study has some limitations. First, since our meta-analysis relies on data from published studies only, the possibility of a publication bias cannot be ignored. In other words, other studies with statistically non-significant results may well not have been included, and the observed impact of interventions might therefore be overestimated. We decided not to draw funnel plots because the number of studies in the review that reported each outcome of interest was low; the funnel plot method is not recommended when a meta-analysis covers fewer than 10 studies [35]. Second, only one of the studies reported data on the incidence of fractures following intervention; no conclusion can therefore be drawn for this outcome. We should not be surprised; the studies evaluating interventions designed to improve osteoporosis screening and treatment typically had a follow-up duration of 1 year or less, a period that is clearly insufficient for assessing the impact any interventions might have on the fracture rate, for it may take years until significant changes in this outcome can be detected [1]. Furthermore, the impact of intervention characteristics could not be investigated using meta-regression models because of the small number of studies included in the review. Finally, the external validity of some interventions included in this review, such as EMR prompts, may be low since these interventions require specialized resources and may therefore not be implemented in other contexts.

In summary, our results suggest that multifaceted interventions targeting high-risk patients and their primary care providers may be effective in improving the management of osteoporosis, but the improvement is often modest, particularly for non-high-risk patients. Since most interventions included in the present systematic review targeted only primary care physicians and their patients, it may therefore be worthwhile if future research dealt with developing and evaluating more intensive multidisciplinary interventions that target various health professionals such as community pharmacists, physician assistants, and nurse practitioners in addition to primary care physicians, in order to insure continuity of care.