Background This study investigated the impact of the Healthy Housing Programme in reducing acute hospitalisations in South Auckland, New Zealand. The programme involved house modifications to reduce overcrowding, insulation and ventilation improvements, and health and social service assessments, referrals and linkages.
Methods An intervention evaluation was used. Participants in the programme were considered cases following their house's intervention and counterfactuals/controls prior to the intervention. Rigorous age-censoring was used to construct a case-counterfactual comparison. 9736 residents of 3410 homes were involved in the programme from September 2001 to December 2007. All lived in areas of relative deprivation (NZDep01=decile 10) and almost all self-identified as Pacific ethnic group. The main outcome measure was acute hospitalisation rates before, during and after a health and housing intervention. Hospital data were gathered from July 1999 to January 2009.
Results In the post-intervention group, people aged 5–34 years had a HR of 0.77 (95% CI 0.70 to 0.85) for acute hospitalisations compared to the counterfactual (pre-intervention). For children aged 0–4 years the HR was 0.89 (95% CI 0.79 to 0.99); a non-significant increase occurred in adults aged 35 years plus. When the causes of hospitalisation were restricted to those related to housing, further falls in the HRs were seen: 0.88 (95% CI 0.74 to 1.05) for 0–4 year olds and 0.73 (95% CI 0.58 to 0.91) for 5–34 year olds.
Conclusion A package of care that addresses housing conditions that impact on health and improves access to health and social services is associated with a reduced acute hospitalisation rate for 0–34 year olds.
- Health service use
- housing and health
Statistics from Altmetric.com
Overcrowding and sub-standard housing conditions can have a significant impact on the health of their residents.1–3 In New Zealand, the most powerful policy changing evidence for the effect of housing on health was provided by work examining a type-B meningococcal disease epidemic in New Zealand. This study3 demonstrated a significant correlation between household crowding and risk of meningococcal disease. One of the policy responses to the epidemic involved a search for a vaccine; another looked towards improving housing stock. The Healthy Housing Programme described herein was a direct consequence of this work. The benefit of improving the warmth of homes on health has also been demonstrated in a recent randomised trial in New Zealand.4
The Healthy Housing Programme has been operating in Counties Manukau District Health Board (CMDHB—an administratively defined health population in South Auckland) since 2001. It is a joint initiative between Housing New Zealand Corporation (HNZC—the provider of government-funded housing) and Counties Manukau, Auckland, Hutt Valley and Northland District Health Boards. A full description is available,5 but, briefly, the Healthy Housing Programme focuses on the specific needs of participating families—all tenants living in rented HNZC homes. There are three related dimensions to the intervention:
Health: aimed at improving tenant access to healthcare services in order to improve health outcomes.
Housing: aimed at reducing the risk of housing related health issues, such as an extension to the house, a transfer to a larger home, housing design improvements or creation of healthy environments, including insulation and ventilation.
Social: a joint intervention that identified social or welfare issues and provided a link to the appropriate social service agencies.
After an area was selected for the programme, all HNZC houses in that area were assessed. A key part of the Healthy Housing Programme is the joint assessment undertaken by an area co-coordinator (housing) and a public health nurse (health). The Healthy Housing joint assessment model has since been adapted for use by other energy efficiency retrofitting insulation projects in New Zealand. From the joint assessment model and home visit, a joint action plan was developed to place the family at the heart of the solution. The team worked with each family to enable implementation of the agreed actions. Interventions included educating families about health risks, referrals to local health providers, making the house dryer and warmer by installing insulation, modifying houses to address health and disability needs, transferring families to other houses to address overcrowding or, in some instances, increasing the number of bedrooms in the house.
By December 2007, 5009 families had been assessed and 3629 enrolled in the CMDHB area. Every participating household received the health and social service interventions; almost all had insulation, heating or ventilation modifications (97%). Interventions to reduce overcrowding (building extensions, transfers) featured for 13% of households. The facilitation and linkage to health and social services work was not ongoing—all work was completed by the intervention completion date.
While not designed as a randomised study, external evaluations of the programme have consistently found improvements in self-rated health, self-esteem, use of primary care and possible reductions in the use of secondary care.6 7 This study examined the impact that the programme had on acute hospital admissions to the local hospital (Middlemore) over the period 2001–2009.
Housing New Zealand tenants living in houses in the suburbs of Mangere, Manurewa or Otara in CMDHB during the time that a house was first assessed formed the study population. Subsequent births in these houses were also followed. Participants were originally enrolled in the study at the time of the initial joint assessment intervention at their address with data captured at this point. A small number of individuals born at that address after its enrolment in the study were also included. All residents of houses enrolled from June 2001 to December 2007 were eligible to be included along with their hospitalisation data covering the period July 1999 to January 2009. Of the 12 644 eligible, 9736 (77%) were able to be linked to the National Health Index (NHI) and, thus, were able to be included in the study. NHI numbers could not be assigned to all eligible participants due to difficulties in data linkage or due to the individual not appearing in hospital records (inpatient, outpatient or emergency care) at any time during the study period.
The population was almost exclusively Pacific ethnic group and all were in the 10% most deprived areas in New Zealand (as measured by NZDep01, the New Zealand Deprivation Index for 2001, a Census-based socioeconomic scoring system for small areas in New Zealand).8 Pacific people are a diverse group of New Zealanders who identify with one or more Pacific ethnicities; the largest in Counties Manukau being Samoan, Cook Islander, Tongan and Niuean.
Coded acute medical-surgical hospitalisations (excluding elective and planned admissions, maternity, mental health and health of older people) to the only local secondary and tertiary hospital (Middlemore) or death recorded in national statistics were used as outcomes. Middlemore is the sole acute care hospital provider for CMDHB and provides about 90% of such care for the CMDHB population. Exceptions include cardiothoracic surgery, neurosurgery and oncology care—services unlikely to be affected by a housing and health intervention. The service mix provided by the hospital has remained constant for the last decade. The standard definition of a hospital discharge in New Zealand includes all patients treated in hospital for 3 hours or more so that all day cases and the more complex casualty/emergency care cases (not admitted to a hospital ward) were included as outcomes in our analysis.
In addition to all acute hospitalisations, a subset of outcomes was developed for this study to measure the impact of housing on health outcomes as no available indicator of ‘housing-related’ health outcomes could be found. ‘Housing-related’ potentially avoidable hospitalisations (HRPAHs) consisted of a subset from the ‘potentially avoidable hospitalisation’ set of hospital diagnosis codes9 previously selected by expert opinion. HRPAHs include hospitalisations (principal diagnosis only) of respiratory or infectious diseases (see table 1) where a strong causal link between the housing intervention and the illness could be postulated through reducing overcrowding or improving ambient temperature in the house.
Exposure status was grouped into three categories depending on the dates in which the housing intervention either started or was finished on their residence:
‘pre’—beginning of study surveillance period (1 July 1999 or birth date if after this date) until first contact with healthy housing programme personnel (joint assessment date).
‘during’—time from first contact with healthy housing programme (joint assessment date or birth date if after this date) until all planned interventions were completed (or death if earlier than this date).
‘post’—time from intervention completion date (ie, all planned interventions completed), or birth date if after this date, until end of study surveillance period (30 January 2009 or death if earlier than this date).
Housing residence was only measured once at joint assessment date (or birth date for those entering the cohort after the joint assessment date). We assumed that participants were resident in CMDHB throughout the study period and continued to reside in their modified houses. We were only able to identify six houses modified from June 2001 to December 2007 that no longer had the tenants that were resident at the joint assessment in occupation. This gives a retention rate of 99.7% by January 2009. We further assumed that they were resident in CMDHB prior to the joint assessment date so that their hospital admissions would have been recorded locally. Healthy Housing Programme tenants had a median 4.8 year occupancy in their residence prior to the joint assessment date (and may have had tenancies prior to that at a different address in CMDHB).
The Cox proportional hazard model was used with the Andersen-Gill counting method to estimate crude and adjusted HRs associated with the exposure status in all three categories.10 Interval time period between hospital admission (or death) was measured using the Andersen-Gill method, which extends the Cox proportional hazard model to allow the effects of exposures on multiple hospital admission events to be assessed. R11 and the survival package12 were used to estimate treatment effects on the hazard of hospital admission (or death) accounting for within-person clustering of outcomes. Further clustering of outcomes, beyond the individual, were investigated by household. Time was recorded using the Anderson-Gill method from initiation of the exposure period to first hospital admission, then from hospital discharge to next hospital admission or end of each exposure period. Individuals were considered censored if they reached the end of an exposure period or if their age category changed without a hospital admission or death occurring. Gender and age were force fit into models. Subjects with unknown gender were excluded from final models; however, no other information was missing. Age was classified in 1-year categories until 5 years with 2-year intervals thereafter except between 20 and 34 years and 35 to 45 years where HRs for hospital admission were similar. Therefore, these middle age groups were aggregated. Period effects were tested as a confounder if inclusion resulted in a change to the β coefficient of the treatment effect (HR) estimate by 10% or more. Period was classified in 2-year periods as a dummy variable based on when the joint assessment intervention was initiated (joint assessment date). Both period and age group were modelled as categorical variables. Standard errors for effect estimates were adjusted for within individual correlation of survival times using an approximate jack-knife procedure. Competing models were assessed by using log-likelihood values. All work was carried out on anonymised data and only aggregated data were reported.
A total of 9736 individuals were included in the analysis (table 2) with 25 659 person-years of post-intervention data, 5190 person-years of ‘during’ data and 42 704 person-years of pre-intervention data available for analysis. Altogether, 95 deaths occurred during the period of observation. The study population was weighted in favour of younger age groups. Period effects or clustering by household did not appreciably change either the point or interval estimate of the housing intervention. All model results are thus presented with adjustment for age, gender and clustering of outcomes within individuals only.
Age was defined at the beginning of the age category within the exposure period. Every time an individual joined a new age category, or changed intervention state, their age changed to the beginning of that category.
The allocation of people into the exposure groups is shown in table 3 with the ‘pre-intervention’ category forming the counterfactual comparison group. As expected, many more individuals entered the study in the 0–11 month age group during the pre-intervention phase than the during or post-intervention phases. Median follow-up in the three exposure groups was 4.9 years (pre), 4 months (during) and 2.3 years (post).
A significant treatment by age interaction was observed so that an average treatment effect could not be quantified. Therefore, the population was divided into three age categories (0–4 , 5–34 and over 35 years) and hazard of hospitalisation assessed for these groups (table 4). Within each group the probability of hospital admission was strongly influenced by age with a steep fall from ages 0–4 years, smaller falls to age 19 years, stable 20–45 years and rises thereafter. For 0–4 year olds, acute hospitalisations reduced by 11% post-intervention (HR 0.89, 95% CI 0.79 to 0.99), while admissions among 5–34 year olds reduced by 23% (HR 0.77, 95% CI 0.70 to 0.85). Hospitalisations among adults aged 35 years and over did not change significantly after the intervention (HR 1.04, 95% CI 0.95 to 1.15). As expected, most of the ‘during intervention’ compared to ‘pre-intervention’ effects were null.
Similar analyses were carried out for HRPAHs. These are summarised in table 5, showing lower HRs than all acute hospitalisations for the younger age groups but higher for the 35 years and over group. The smaller numbers of outcomes are reflected in wider CIs. The 5–34 year old age group remains significantly reduced (a 27% reduction) while the 35 years and over group shows a significant 31% increase in acute hospitalisations for these conditions.
Our study provides evidence that a combined intervention to improve health and social service access, and housing conditions—overcrowding and improving warmth and ventilation—is associated with a reduced rate of acute admission to hospital in those aged 0–34 years. It is based on a large number of houses and people (9736 residents of 3410 homes) with a median of 2.3 years of post-intervention data. The Healthy Housing Programme is associated with a reduction in all acute hospitalisations by 11% in 0–4 year olds and 23% in 5–34 year olds in the study population. For housing-related hospitalisations the falls were 12% and 27%, respectively.
Strengths and weaknesses of the study
Although our analysis was not a randomised trial, we used a within-person, crossover design from the hospitalisation data for the residents of the houses prior to their intervention so that the effect of the intervention could be estimated. Carrying out a crossover study in this way avoids the use of controls drawn from the population, which are subject to unknown biases. The use of data independent of the programme itself (routinely collected hospital data) adds further face validity and confidence in the results. The study has a large sample size allowing relatively rare events (acute hospitalisations) to be studied, and enabling the use of age and interval-censoring to be carried out to control for age effects that might otherwise dominate a simple before/after comparison. The results are concordant with the formal evaluations of the programme6 7 and provide further impetus for its continuation and expansion to equivalent households.
Our statistical model censored individuals who reached the end of their age category. For example, children aged 3 years in post-intervention houses were compared to 3 year olds in pre-intervention houses, then when they turned 4 they were compared with 4 year olds in pre-intervention houses, and so forth. Using this approach of changes over time in hospital admission rates, whether through specific interventions or caused by ageing, should have equal effects on both groups. No period effects were noted in the models used. In a separate analysis across the District Health Board a 1.6% increase per annum in age-standardised acute hospitalisation rates occurred over the study period. To the extent that any residual period effects remain they likely would have acted to reduce the size of the effect we have reported. We had expected that increased duration of living in a modified house would increase the size of the effect seen but this did not prove to be the case (although more long-term studies would be needed to confirm this).
Not all eligible Healthy Housing recipients were included in the study resulting from our inability to link data to the NHI code used to identify individuals in New Zealand health records (77% were linked).13 However, only people that were able to be linked were included in the counterfactual, and the chance of being linked should not vary for the pre-intervention or post-intervention groups, so biases should be minimised although cannot be ruled out. Age showed a strong ‘U’ shaped risk, higher at younger ages, plateauing in the 20s to 40s and increasing again in middle and older age groups. We controlled for age in yearly and 2-yearly increments for 0–19 and 45+ year age groups; however, residual confounding by age cannot be ruled out.
The unit of analysis here was persons living in the house at the time the joint assessment was performed. No attempt was made to track individual's addresses forward or back from that date—effectively an intention-to-treat design. Although this potentially may result in misclassification of both exposure and outcome, any bias that this might cause should affect the counterfactual group at the same level as the intervention group and tend to reduce the apparent size of any impact. Post-intervention, the programme reports little movement of people from their modified houses (a remarkable 99.7% retention) so the main concern is in the pre-intervention period. A median tenancy of 4.8 years prior to the joint assessment date indicates that most participants would have had their data recorded in the event of a hospital admission. To the extent that we may be undercounting hospitalisations for people who moved into the district just prior to participating in the programme, the study will be conservative, having lower hospitalisation counts in the counterfactual/control group, biasing towards a null association. In addition, babies born into the household after the joint assessment date were specifically sought to provide a comparison group for the youngest age groups that would not have otherwise had appropriate ‘post’ comparators.
Relation to other studies
To our knowledge this is the first study to show that general housing improvements linked with improved health and social service access is associated with reduced need for acute hospital treatment. In their reviews in 2000 and 2009,14 15 Thomson et al did not identify any controlled trials that showed reductions in this outcome. Howden-Chapman et al showed an OR of 0.90 for hospital admissions (95% CI 0.59 to 1.37) and a larger, although still non-significant, reduction in the need for hospital treatment due to respiratory conditions (OR 0.53, 95% CI 0.22 to 1.29) following the installation of insulation in houses.4 The findings in this study support the conclusions of the meningococcal case-control study, which highlighted an association between household crowding and this life-threatening infectious disease.3
Unanswered questions and future research
In addition to examining the impact on overall acute hospitalisations, a subset of potentially avoidable hospitalisations (PAH) thought to be housing-related were selected based on the principal diagnosis of each event. This sub-setting of the outcome data showed an increased point estimate of the effect size (albeit with wider CIs due to the smaller number of events) lending some confidence to the measures used. The PAH rubric10 is a development of similar listing from other researchers that are used routinely in many countries. For example, a set of ambulatory sensitive hospitalisations (ASHs) is used by the New Zealand Ministry of Health to monitor health system performance.16 Less well-documented are attempts to relate housing conditions to specific hospital events. We were unable to find such a list in the literature; hence, our construction of the HRPAH measure used here. The logic underlying this measure consisted of a hypothesis that infectious and respiratory diseases are promoted by damp, cold and overcrowded living conditions.1–4 6 11 The diagnoses selected were restricted to those present in PAH—for example, this led to the non-inclusion of viral pneumonia (not included as an ASH or PAH due to the lack of treatment possibilities) where starting a list from scratch one might have included it. For children in this cohort, the HRPAH conditions made up around 40% of the acute hospitalisations, and for adults, around 15%.
Our a priori assumption was that the programme would be most effective in younger children who have the highest rates of hospital admission for infectious disease. In the event the highest rates were in the older, 5–34 year, age group. This appears to relate to respiratory infections, asthma and cellulitis, but further work is needed on this.
The apparent increase in acute hospitalisation in older adults—significant for the HRPAH subset—requires further exploration. Our working hypothesis would be that it relates to the health intervention offered to participants as part of the programme. Adult participants are better connected to the health system as a result of the programme.6 7 While the specific health status of the residents was not included in this study, we would note that Pacific people aged 35 years and over in the CMDHB area have a rate of obesity of over 80% and 23% are estimated to have diabetes (and by age 55 years, 40% are diagnosed with diabetes).17 Alternative explanations include residual confounding by age despite our attempts to control for it.
We observed that an improvement in housing conditions combined with a health and social service linkage intervention was associated with reduced acute hospitalisation rates in a socio-economically deprived sample of Pacific residents of New Zealand. This is one of the largest health and housing community trials reported. Combined with earlier reports, such results increase the likelihood that when health and social service access is improved, overcrowding is reduced and adequate ventilation and warmth is provided tangible health benefits result.
As a significant government-funded intervention the Healthy Housing Programme has been active for nearly a decade now. It has undergone internal and external evaluation to demonstrate that it is possible to work across agencies/sectors to directly improve the living conditions of the most deprived people in our society.6 7 Multiple interventions can have additive effects. Of particular note in this group of low-income tenants is the housing stability evidenced by the high retention rate. The increased sense of empowerment noted by tenants in the evaluations and improved comfort they reported in their homes may result in improved family functioning and cohesion leading to an increased sense of social well-being.6 This improved self-esteem is likely to impact on many aspects of the families' lives well beyond any effects seen in the healthcare sector.
What is already known on this subject
Overcrowding and sub-standard housing conditions impact on the health of their residents.
Interventions to improve housing conditions can generate health improvements.
What this study adds
Improving housing conditions along with providing health and social service linkages is associated with a significant reduction in acute hospitalisation for 0–34 year olds. This is the first study to show that a housing-based intervention can improve health to the extent of reducing acute hospitalisations and is one of the largest intervention studies reported in this area.
Health inequalities are a major issue in most health systems. Well-tested interventions to reduce health inequalities are relatively few with many of the root causes for health inequalities lying outside the traditional healthcare sector. The Healthy Housing Programme demonstrates that it is possible for health services to work across sectors to improve the health of the population directly. This study provides more impetus for healthcare commissioners to look for opportunities for intersectoral action with their housing counterparts.
The authors would like to thank the all the participants in the Healthy Housing Programme for their willing participation; and the programme staff from Counties Manukau District Health Board and Housing New Zealand who went the extra mile and assisted the data gathering and coding for programme evaluation well beyond their normal duties.
See Commentary, p 598
Linked article 132407.
Competing interests No financial interests are involved. All authors have been employed by the two organisations responsible for funding the Healthy Housing Programme. This work and any views expressed are solely those of the authors and not of their employing agencies. No editorial control came from either organisation.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.