Poor housing conditions and the risk of falls: Previous evidence

Previous evidence has highlighted the importance of the home environment for falls prevention, though with varying emphases. Systematic reviews have demonstrated that home safety interventions, often delivered as part of multifactorial programmes, can reduce the prevalence of falls among older adults living in the community [17–19]. These interventions usually involve relevant professionals assessing home environments and identifying hazards, and then recommending modifications such as fixing loose carpets, repairing poor stair conditions, dealing with slippery floors, improving lighting, and/or clearing cluttered pathways. While these approaches seem promising, they typically adopt a holistic model in which multiple risk factors are addressed simultaneously, making it difficult to isolate the specific role of housing quality in falls prevention.

Evidence linking poor housing conditions directly to falls is more limited. International and national studies have increasingly recognised the importance of housing quality [20]. Lack of space, for instance, may increase tripping hazards, obstruct visibility, and movement [21], while low indoor temperatures have been associated with poorer physical functioning and a greater risk of accidents [22]. Policy initiatives, such as the “Boiler on Prescription” trial in Sunderland, and other areas suggest that improving indoor temperatures can reduce falls and hospital admissions [23], although robust evaluations remain scarce.

Despite these attempts, most of the existing evidence has prioritised housing adaptations over housing quality. Adaptation-focused interventions have been demonstrated to have preventative potential, but they may not address the more systemic risks arising from substandard housing conditions. Moreover, much of the available economic evidence on fall costs is confined to hospital or residential settings, with limited attention to community-dwelling. In Scotland, community-based falls in older adults were estimated to cost over £470 million in healthcare and social care expenditure [24], underscoring the urgency of preventative strategies that extend beyond clinical and behavioural domains. Overall, while environmental interventions are recognised as an important component of falls prevention, research evidence remains fragmented and underdeveloped.

Falls

In ELSA, respondents were asked whether they had fallen since their last interview: “Have you fallen down in the last two years (for any reason)?” This was coded as a binary variable (0 = No, 1 = Yes). Those who reported a fall were further asked: “In any of these falls, did you injure yourself seriously enough to need medical treatment?” This was also coded as a binary variable (1 = Yes, 0 = No). Information on the specific type of medical service was not available.

Housing

ELSA collects detailed information on housing characteristics and quality. These include the number of household members, number of rooms, and whether adaptations had been installed (e.g., handrails, automatic doors, and alerting devices). Self-reported housing problems were also recorded, including issues with lack of space, general disrepair, cold indoor temperatures, lack of light, noise, condensation, water leaks from roofs or windows, electrical or plumbing faults, infestations, and rising damp. Each problem was coded as 1 if present and 0 if absent. We also included fuel poverty as a potential factor increasing risk of falls. Here (due to data limitation on energy consumption and housing insulation levels), we defined fuel poverty as those households that spend 10% or more of their income on energy (gas and electricity) bills. The latter was coded 0 if individuals are not living in fuel poverty and 1 if they are.

Socio-demographic characteristics, care needs, and health status

Socio-demographic information included gender (0 = male, 1 = female), marital status (0 = single, 1 = married), long-standing illness (0 = no, 1 = yes), living alone (0 = no, 1 = yes), age, and self-rated health (0 = fair/poor/very poor, 1 = good/very good). Care needs are also found as the need for support with Activities of Daily Living (ADLs) and Instrumental Activities of Daily Living (IADLs). ADLs include support with going to the toilet, eating, bathing, personal care, etc. and IADLs include the need for support with tasks such as shopping, making calls, taking medication, and managing money, among other things. Once again, the need for support was coded with a 1, and 0 otherwise.

Cost of healthcare and social care services

We obtained the cost of formal care and healthcare services using PSSRU and NHS unit costs for the year 2023/2024 [26–27]. Cost of NHS 111 services were not available, but an estimation was used based on [28]. Older people (65+) population size for England was obtained from Census 2021 [29].

Cross-lagged panel model

To tackle reverse causality, we estimate the effect of housing conditions measured at time t–1 on the risk of falls at time t. This is a standard, simple and effective way to rule out the possibility of a bidirectional relationship as the past can influence the future, but not the other way around [30]. The model controls for socio-demographic characteristics, long-standing illness, and history of previous falls at t–1. The main specification is given by:

(1)

Where is a binary indicator of whether individual experienced a fall at time , represents socio-demographic, health-related factors, previous falls, and local deprivation levels (quintiles of the Index of Multiple Deprivation). captures reported housing quality issues as binary variables), and represents the error term. We have performed, in the first instance, a panel-data (Random Effects) probit regression model, with clustered and robust standard errors.

Heckman selection model

To account for potential bias arising from an association between poor housing conditions and mortality or attrition between waves, we estimated a two-step Heckman correction model [31]. The first-stage selection equation predicts the probability of an individual remaining in the sample at time t, using auxiliary variables correlated with sample retention but not directly with falls outcomes. We tested several instruments, including smoking status, number of completed survey waves, incidence of heart attack, exposure to pollution, enjoyment of social company, receipt of bequest (£50k), receipt of attendance allowance, and combinations thereof. The final specification employed smoking status (0 = no, 1 = yes) and number of survey waves participated in (ranging from 1 to 10), which satisfied relevance and exclusion criteria.

The outcome (second stage) equation is then estimated as:

(2)

Where is the inverse Mills ratio (IMR) derived from the first-stage selection equation

This modelling framework allows us to (i) reduce concerns of reverse causality, by using lagged housing measures, and (ii) correct for potential sample selection bias. Together, these approaches provide a more robust estimate of the causal impact of poor housing quality on falls among older adults in England. To aid interpretation, we report marginal effects derived from probit models for housing conditions. Unlike raw probit coefficients, which represent changes in the unobserved latent index, marginal effects express the change in the probability of experiencing a fall associated with a one-unit change in the explanatory variable. These are calculated as the product of the estimated coefficient and the standard normal density evaluated at the mean of covariates. Stata 18 [32] was used to perform all analyses.

Decision tree model

We developed a decision-tree model to estimate the healthcare and (formal) care consequences and costs of falls among older people (65+) living in poor housing conditions (defined here as lack of space, poor lighting, and plumbing/electrical issues). The model follows a cohort of older adults through mutually exclusive care pathways triggered by a new fall (e.g., calling an ambulance and attending A&E, GP attendance with or without referral, no healthcare use), which assigns probabilities to transitions along the tree and attaches unit costs to each service use. For each terminal node, we also included the subsequent likelihood of short-term and long-term community or residential care. Total annual costs were obtained by aggregating costs across pathways, scaled up to the population level using ONS population estimates for England taken from Census 2021 [29]. Fig 1 presents a visual representation of the developed decision tree:

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Decision tree pathways.

https://doi.org/10.1371/journal.pone.0353755.g001

Developing pathways and assumptions

The decision model was mostly designed in a deterministic manner. In order to estimate the additional cost associated to falls caused by poor quality housing, we considered all older people in the sample, and we estimated the percentage change (i.e., the additional effect) in the risk of falls associated with living in poor housing conditions. To do this, we applied a multinomial probit Heckman selection model (the only non-deterministic part of the model) with three main outcomes: (i) having a fall at time t; (ii) not having a fall at t but having had a previous fall at t-1; and (iii) not having experienced a fall. The model took the same form as in Equation 2 above.

The estimated coefficient β was converted into a marginal effect, allowing interpretation as the percentage change in the probability of falls associated with living in poor housing, with good housing as the reference category.

Probability for downstream events in the tree (e.g., calling an ambulance, of attending A&E, of sustaining a hip fracture, referral to falls services or OT/PT) were drawn from the published literature and policy documents (see Table 1). Unit costs for ambulance calls, A&E attendance, hospital treatment (hip fracture, other fracture, laceration), home care and institutional care were taken from established UK sources and are presented in Table 2. Some assumptions were also required due to the lack of available data.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Parameters and assumptions in decision tree model.

https://doi.org/10.1371/journal.pone.0353755.t001

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Unit costs.

https://doi.org/10.1371/journal.pone.0353755.t002

The decision tree begins with the subpopulation of older people living in poor housing and branches into: those who experience a new fall in the year and those who do not. For fallers, the tree routes individuals through possible immediate responses (call ambulance, A&E attendance, and categorised clinical outcomes; call 111; attend GP; no healthcare use), then into consequent care states (no further care, short-term home care, long-term home care, short-term institutional care, long-term institutional care). For each branch, the estimated cost is computed as the product of the branch probability and summed unit costs along that branch. Formally, the expected annual cost for branch j is:

(3)

and

(4)

Total costs were divided into two categories, healthcare and social care related’, and are presented in 2023 prices. Estimations of the latter include local authority-funded and self-funded services.

It is also worth noting that the decision tree model was not populated based on a systematic review. We did ensure, however, that the papers used were relevant (e.g., based on UK data as much as possible) and based on representative samples (as well as other PRISMA-relevant indicators).

Triangulation

To triangulate our cost estimations, we compared results with published estimates of healthcare and formal care costs associated with falls. In particular, we used Craig et al. (2013) [24] which provides aggregate health and social care cost estimates for older adults experiencing falls in Scotland. To make their findings comparable, we adjusted Craig et al.’s figures for inflation and scaled them by population size (to reflect England rather than Scotland, and to account for differences in the 65 + population). We also took account of the fact that their figures relate to all falls in older people, rather than those specifically attributable to poor housing.

One of the main methodological differences from Craig’s work (in addition to those already mentioned above) is that we estimated the cost of institutional care based on the distribution of length of stay reported by Forder and Fernandez (2011) [33], whereas Craig simplified this by applying the median length of stay. We also included additional healthcare services such as NHS 111 or referrals to an Occupational Therapist by a GP, services that were not considered by previous works. On the other hand, due to the availability of administrative data, they were able to estimate re-admissions, something which we could not consider (although this may be a minor issue given the time cap).

Falls prevalence

We started the analysis by estimating the prevalence of falls among older people living in the community in England using the English Longitudinal Study of Ageing – ELSA (using waves 1–10, covering the period 2002–2023). Approximately a third of older individuals (65 years or older) experience one or more falls a year. One in ten older people reported visiting healthcare services due to a fall. These results are consistent with estimations provided by the Office for Health Improvement & Disparities [1], see Fig 2 for more details.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Prevalence of falls among older people living in the community (ELSA waves 1-10).

https://doi.org/10.1371/journal.pone.0353755.g002

Poor housing conditions associated to an increased risk of fall

Using a cross-time-lag approach within a Heckman selection model, our results show that those living in homes that have electrical or plumbing issues (β = 0.274; p-value = 0.011) as well as those with lack of space (=0.084; p-value = 0.056) are more likely to experience a fall. Similarly, we found that older people living in homes which are ‘too dark or without enough light’ (β = 0.236; p-value = 0.011) and again have ‘electrical wiring or plumbing issues’ (β = 0.197; p-value = 0.100), are more likely to experience a fall and visit a healthcare service as a consequence, which may be interpreted as a more serious fall. In this analysis we have controlled for: previous fall, age, gender, marital status, education, long-standing illness, wheelchair use, health status, number of ADLs/IADLs, receiving informal care, and tenure.

Due to the collinearity between tenure (privately renting, social housing, and owner occupied) and housing condition, this variable was excluded as sensitivity analysis (column 7 and column 10 in Table 3). As expected, we see an increase in the effect of housing conditions when tenure is removed from the model of falls (β = 0.278; p-value = 0.010 for electrical/plumbing issues and β = 0.096; p-value = 0.031 for lack of space) and to the model of falls with healthcare use (β = 0.197; p-value = 0.100 for electrical/plumbing issues and β = 0.251; p-value = 0.006 for lack of space). Heckman selection model selection bias test indicated that there is no evidence of selection bias (Prob > chi2 = 0.5830). Therefore, we also reported cross-lag only models, for which we obtained very similar results. Other housing-quality issues, such as general disrepair, cold indoor temperatures, noise, condensation, water leaks from roofs or windows, pest infestations, and rising damp, were also tested in the analysis but were not found to be statistically significant. Local deprivation levels were also tested but were not found to be statistically significant.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Regression models.

https://doi.org/10.1371/journal.pone.0353755.t003

Table 3 reports on the different specifications and methods performed. For an easier interpretation, we have also included the effect of housing conditions as percentage points. For instance, reporting plumbing/electrical issues increases the risk of fall by 10.8%. Area-level deprivation (IMD quintiles) was also included in the model as an additional control, but it was not statistically significant and was therefore excluded from the final specification.

In the Heckman selection models, smoking status and number of waves were incorporated as auxiliary variables within the selection equation to strengthen model identification and improve overall model fit. We acknowledge, however, that the appropriateness of smoking status as an exclusion restriction may be questioned, given the established physiological relationship between smoking, frailty, and risk of falls. Consequently, smoking may plausibly exert a direct effect on the outcome beyond its association with sample selection. To examine the robustness of the findings to this assumption, robustness analyses were undertaken excluding smoking status from the selection equation. As shown in Table 4, the results were consistent, indicating that the principal findings were robust to alternative model specifications.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Robustness analysis.

https://doi.org/10.1371/journal.pone.0353755.t004

The health and social care costs of falling due to poor housing

Combining branch likelihood with unit costs, we estimated that £1.2 billion annual costs are attributable to falls among older adults living in poor housing. The expected average cost per older people with a fall living in poor housing was £869. Healthcare costs were estimated to be £440 million (£313 per older people), and formal care costs were estimated at £778 million (£556 per older people), both in 2023 prices.

When we compared our estimations to Craig et al.’s work [24], adjusting the Craig figures for inflation and population size, and accounting for the fact that Craig et al. [24] report costs for all older people rather than costs attributable to poor housing, our results were fairly similar in magnitude to the Craig estimate. Adjusted healthcare costs were in our case £1.7 billion (36% of the total) versus £2.4 billion (40% of the total), formal care costs were £2.9 billion (64% of the total) versus £3.5 billion (60% of the total), and a total cost of £4.6 billion versus £5.9 billion. Where differences arose, they were explainable by (i) our focus on a housing-specific subpopulation, and (ii) differences in assumptions about transition probabilities, length of stay, and unit costs. This triangulation provides additional face validity for the modelled results.

We also triangulated our results from a report published by Age UK [45] where health costs of falls were estimated for England, obtaining similar figures: £1.7 billion (versus our £2.0 billion). We couldn’t however obtain the methodology used for this estimation, so we adjust only by inflation and by total cost of falls (rather than confined to those caused by housing conditions). We also attempted to compare our results to a macrosimulation model estimating savings over time associated to fixing home conditions and social care service use [46]. Unfortunately, due to being a project on costs comparison, we were not able to compare findings.

Sensitivity analyses were also conducted to assess the robustness of the cost estimates to key parameter assumptions in the decision tree. While results remained consistent across specifications, total cost estimates were sensitive to parameter variation, ranging approximately from £1 billion to £2 billion.

Discussion & policy messages

Using information from the English Longitudinal Study of Ageing, we found that around one in three people aged 65 or older experienced at least one fall each year, and approximately one in ten reported having used healthcare services as a result, which is consistent with previous studies [1]. Our analysis also shows that specific housing-related problems, such as electrical or plumbing issues, lack of space, and inadequate lighting, increase the likelihood of falling and of accessing healthcare services following a fall. These effects remained statistically significant across different specifications and modelling approaches. The mechanisms underlying these relationships could plausibly be: water leaks or uneven flooring caused by plumbing problems increase the likelihood of slipping, exposed wires or cluttered cables create tripping hazards, insufficient light makes it harder to identify obstacles, and cramped living spaces restrict movement and contribute to cluttered or obstructed pathways. While other housing-related factors, such as poor insulation or damp, were not statistically significant in our models, this may reflect self-report limitations or strong collinearity across housing problems rather than an absence of effect.

From a financial perspective, we estimated that £440 million and £778 million in annual health and formal care costs, respectively, can be attributed to falls linked to poor housing conditions. The total estimated costs (£1.2 billion) in our study are comparable in magnitude to those reported by Craig et al. (2013) [25] and Age UK (2010) [45] when adjusted for inflation, population, and methodological scope. Moreover, we found that social care costs represent the largest share of the financial burden, highlighting the potential for housing improvements to alleviate pressures on already stretched care systems.

These findings have important implications for policy and practice. Improving housing quality should be recognised as a key component of falls prevention and healthy ageing strategies, alongside clinical and behavioural interventions. The results also highlight the need for stronger collaboration between housing, health, and care sectors to identify individuals at risk and coordinate timely interventions. This aligns closely with the NHS 10-year plan [47], which emphasises the importance of integrated, place-based approaches to improving population health and addressing the wider determinants of wellbeing. Investing in safer, better-quality housing within communities directly supports this agenda, offering a practical route to reduce inequalities and promote healthy ageing at the local level.

Preventing falls through environmental improvements not only represents a health and safety priority but also a means of achieving broader system efficiencies and social benefits. Importantly, many of the improvements needed, such as better lighting, minor repairs, or decluttering and rearranging furniture, are simple and relatively inexpensive to implement, yet can have a significant impact on safety and wellbeing. In this regard, older adults living in poorer-quality housing (often those in rented housing or deprived areas) may be prioritised for housing assessments and adaptation grants. Investment in housing quality can thus serve as an effective form of prevention, reducing hospital admissions and reliance on long-term care services while supporting older people to live independently for longer.

Limitations

This analysis has several important limitations that should be acknowledged. First, the measure of falls in ELSA is based on self-reports. This may be subject to recall bias or under-reporting due to stigma, memory lapses, or concerns about self-worth, potentially leading to underestimation of fall prevalence. Additionally, ELSA does not provide a breakdown of the type of healthcare used following a fall, so we are unable to distinguish between GP visits, A&E attendances, or walk-in centre consultations. Second, while we took several steps to mitigate potential sources of bias, we recognise that some degree of omitted bias is inevitable in observational research.

Third, the decision tree model used to estimate costs represents a simplified view of the complex pathways through which falls result in healthcare and social care use. While this approach allows for consistent cost estimation, it abstracts from the diversity of real-world trajectories and relies heavily on assumptions and evidence drawn from previous literature. The model also focuses on one-year outcomes and may not capture longer-term consequences of falls. Forth, mental health service use due to anxiety, fear of falling, or other psychological consequences, were not included due to data limitations. Falls can lead to substantial emotional and psychological impacts, potentially resulting in additional service use.

Fifth, the analysis distinguishes between falls associated with poor housing versus other risk factors. However, theoretical differences in these pathways and their downstream consequences have not been fully accounted for. This may contribute to an underestimation of costs. Supportive of this, descriptive analyses in ELSA show that among individuals who experienced a fall, those living in poor housing were slightly more likely to report healthcare use than those in better housing (34.78% vs. 31.15%).

Finally, the accuracy of our cost estimations heavily relies on previous evidence used to populate our decision tree model.

Taken together, these limitations suggest that while our results provide useful estimates of the additional costs of falls associated with poor housing, they are likely to be conservative. Future work may seek to incorporate more detailed information on healthcare use and mental health services, to provide a more comprehensive picture.