The impact of community-based interventions for the older population: a quasi-experimental study of a hip-fracture prevention program in Norway

In Norway, the municipalities have the overall responsibility of residents’ health [14]. The municipalities are obligated by law to promote health, and prevent injuries, accidents and social problems among its citizens. They are also responsible for funding and providing necessary primary care to their residents.² Necessity is defined by the health needs of residents. Municipalities are restricted to allocate services according to health needs and independently of socioeconomic status [23]. Among the primary care services are all social and community health services provided to persons with long-term care (LTC) needs. The LTC services the municipalities finance and provide can be broadly divided into nursing and home-based care services. The LTC sector can be said to be semi-centralized [13]. In the sense that the central government determines the legal bounds of municipalities’ health and care responsibility, while the municipalities have extensive discretion in determining the composition of preventive, long-term, and curative care that best meet the needs of their residents.

LTC expenditures is the largest component of municipal spending [14]. There is a fear that this funding responsibility will become more demanding in the future because of an aging population. This has led many municipalities to focus more on measures that can prevent and postpone care needs. One such measure is interventions to prevent hip fractures. Hip fracture prevention efforts have received a lot of attention in Norway. This is because Norway has one of the highest hip fracture incidence rates in the world [21, 28], and experiencing a hip fracture causes functional decline and increased need for LTC [40, 43]. Osnes et al. [36] and Hektoen et al. [16] are two studies that respectively show the consequences of hip fractures on LTC costs and needs in Norway. Osnes et al. [36] estimate the likelihood for older people living independently in the community of needing LTC after experiencing a fracture to be above 50%. Hektoen et al. [16] calculates the health and care costs in the first year following a hip fracture and find that the greater part of the costs is LTC costs borne by the municipalities. Both studies recommend municipalities to establish hip fracture preventions programs.

In 2007, 15 (of 428) Norwegian municipalities and one district in Bergen implemented the Safe Aging program (in Norwegian: Trygge eldre).³ The program was initiated by the non-profit organization Forum for Injury Prevention (in Norwegian: Skadeforebyggende forum), and the aim of the program was to prevent falls and fall-related injuries among older adults (≥65 years). The program focused especially on reducing hip fractures because of the large associated treatment and quality-of-life costs [40]. Following Lund and Aarö [27], the program used a wide range of measures aimed at changing attitudes, behavior, and the physical and organizational environment. The program was implemented in a non-centralized way and the municipalities could choose which measures to implement and how to design them, the municipalities mainly financed the implemented measures themselves. The project leader of the Safe Aging program worked as a coordinator, and supervised contact persons in all participating municipalities. An important part of the program was to support cooperation between municipalities and voluntary activities. In the end, the municipalities focused on similar measures, e.g. information to employees, older adults and relatives, education for employees, physical exercise, preventive home safety assessment and modification help services, and fall registration [40]. We present an overview of the implemented measures in Table 1.

All participating municipalities implemented some kind of information measures. The information measures included brochures on safety measures, and the benefits of physical exercise to reduce the risk of hip fractures, either distributed to all households in the municipality or via organizations for retirees and other voluntary organizations. Lectures or courses on the same themes were also held in most of the participating municipalities. Five of the participating municipalities also educated their employees on how to reduce fall-related injuries. Nine municipalities implemented some kind of physical exercise for the at risk population. For the reasonably fit older adults, a wide range of activities were on offer: from weight training at gyms, to Thai-Chi, and exercise in swimming pools. For those in need of more help, living at home or in an institution, exercise groups were held in the institutions or in local health centers. Finally, several municipalities offered preventive home safety assessment and modification help services, where focus was on measures to prevent accidents in the home, but also information about the benefits of physical exercise and information about the supply of preventive measures given in the municipality. Seven municipalities implemented systematic registration of falls in both home services and in institutions. Most of the implemented measures have not stopped at the end of the project period, but are ongoing.

To measure hip fracture incidence in Norwegian municipalities we collected hospital admissions data from the Norwegian patient registry (NPR) in the period 1999 to 2014. NPR includes, among other variables, information on diagnosis and procedure codes, age, gender, and patients’ municipality of residence for all in and outpatient stays from 1999 [4].

The applied method requires a complete time series in the dependent variables; we therefore dropped municipalities that were involved in mergers in the study period. In 1999, there were 435 municipalities, and of those 14 municipalities were involved in mergers. In addition, we did not have information on hip fracture rates in the districts of Bergen, and since only one city district in Bergen participated in the program, we excluded Bergen from the analysis. Our sample therefore consisted of 420 municipalities, of which 15 were in the treatment group.

There are several ways to quantify hip fracture incidence using hospital registry data. See for example Omsland et al. [35], who compared different methods using registry data. We followed Øien et al., [46], and used the method recommended in Høiberg et al. [18]. Høiberg et al. [18] drew a random sample of admissions from NPR registered with diagnosis and/or procedure codes related to hip fracture, and analyzed the information contained in hospital records for this sample. They found that defining hip fractures according to a combination of diagnosis and procedure codes gives the best correspondence with actual hip fractures as indicated by hospital records. Thus, we defined hip fractures as all hospital admissions coded with a diagnosis code for femoral fracture (ICD10, S72.0–2) and either procedure code (NOMESCO version 1.14) for treatment of femoral fracture (NFJxy, x = 0–9,y = 0–2) or for replacement of hip joint (NFBxy, x = 0–4, y = 0–2; NFB62).

We used a difference-in-difference (DD) design to evaluate the effects of the interventions on hip fracture rates. It is well known that the causal assumptions behind the DD method rely on a strong common trends assumption, in that for the estimates to reflect the causal impact of the intervention, both the treated units and the controls would have needed to follow common trends in absence of the intervention (Angrist & Pischke 2011). Using a nearest-neighbor matching algorithm developed by Bonander [5], we therefore identified our control municipalities by matching on pre-intervention trends, which should decrease bias under the assumption that there are no other changes in the post-period that affect hip fracture rates in the analyzed units. After matching, we quantified the average treatment effect on the full treatment group (n = 15) in several ways: (1) average post-intervention effects on hip fracture incidence rates per 100.000 person-years, (2) time-varying (dynamic) effects on incidence rates, (3) relative effects (incidence rate ratios) and (4) in the cumulative number of hip fractures prevented during the observed post-period. P-values and confidence intervals were calculated assuming a Poisson distribution on the underlying counts. Cluster-robust standard errors were computed by estimating a design effect based on the intra-cluster correlation coefficient in the matched samples, which works well for controlling the false rejection rate in small sample fixed-effects analyses [32]. We also performed a set of subgroup analyses to test for the moderating effects of age group (above 80 years) and sex, and estimated municipality-specific treatment effects due to the variations in interventions implemented by the different municipalities (Table 1). We summarized the relative effect estimates from the latter in an inverse variance fixed effects meta-analysis to formally test for the presence of heterogeneity in the effects using a Q-test [17]. The analysis was performed in R using the idd and metafor packages [5, 44].

In Table 2, we present the mean and standard deviation of the number of hip fractures per 100,000 inhabitants across age and gender for treatment municipalities and potential control municipalities in the pre-reform year 2006. As expected, we can see that age and gender are important risk factors for hip fracture. The number of hip fractures in the age group 80 years and older was roughly twofold compared to the number in the whole population 65 years and older. Women were roughly two-thirds more likely to experience a hip-fracture than men in our population.

In 2006, the hip fracture incidence rate at ages 65 and older was fairly similar in treatment municipalities and potential control municipalities. However, the incidence rate at ages 80 years and older, for both genders, was lower in treated municipalities compared to the average of the potential control municipalities. This is somewhat surprising since one would believe that municipalities with high hip fracture incidence rates would to a greater extent implement hip fracture prevention interventions. On the other hand, it could be the case that treatment municipalities, on average, were facing a higher growth in hip fracture incidence rates, and therefore implemented prevention activities to curb this trend. As explained in the previous section, our method deals with differences in levels and pre-intervention trends in the hip fracture incidence rates between treatment municipalities and matched control municipalities, such that these differences should not bias our results.

We present the results from the main analysis in the top left panel in Fig. 1. The matched control municipalities followed similar hip fracture trends as the treated municipalities, which speaks for the validity of the effect estimates. As can be seen in Fig. 1, the units did not diverge significantly in the post-period, suggesting that the reductions we see in the treated municipalities also occurred among the controls. Hence, the average intervention effect estimates for the full treatment sample are non-significant and close to zero. The point estimates and confidence intervals are presented in Row 1, Table 3.

As in the main analysis, the matched controls followed similar hip fracture trends as in the treated units across the entire pre-period (Fig. 1; Table 2). Overall, the results from the subgroup analysis were similar to the main analysis in that we found no statistically significant effects for the age group 80 years and above, or in the sex-stratified analyses.

We present the results from a meta-analysis of municipality-specific models in Fig. 2. Overall, the effects were clustered around the null, with a non-significant (weighted) average effect of − 11%. There was no significant evidence of heterogeneity, according to a Q-test (Q(df = 14) = 7.8, p-val = 0.9).

There are some limitations to our study that should be mentioned. First, the non-random allocation of intervention municipalities poses a threat to the internal validity of the results. However, our quasi-experimental design specifically matches on pre-intervention trends to find controls that (at least in the pre-period) follow the same trends on the outcome as the treatment units, which decreases the risk of bias. Still, while we are unaware of any other large-scale programs or interventions implemented at the same time, other post-intervention changes that are unrelated to the intervention would bias the results if they affect hip fracture rates in the treatment or control units. Another limitation is that we were unable to measure the actual fidelity of the implementation to that presented in the final report [40], and therefore cannot, without speculation, identify whether the lack of evidence of an effect is due to theoretical failures in the design of the intervention or failures in the implementation of the planned activities. We were not present during the implementation of the interventions, and could therefore not conduct any formal process evaluation during the implementation phase. To our knowledge, no critical assessment of the program theory or its implementation has been conducted. While process evaluation can be done retrospectively [9], the data collection required for a comprehensive analysis of the implementation process was beyond the scope of our current study. We strongly recommend that process evaluation is conducted alongside similar interventions in the future One last limitation we would like to mention is that in aggregated data analysis it is often the case that effect sizes are imprecisely estimated. This is because effects were evaluated for the whole target population, and not only for the population at risk where the treatment is actually expected to have an impact. Therefore, in community trials there is often not enough precision to detect small effects. This is another potential explanation for why the evidence of community interventions is weak.