A cross-sectional design was used. Randomisation took place on two levels. Firstly, a random and weighted sample of 15 nursing homes was obtained from a list of all Dutch nursing homes 22. These 15 nursing homes represent 5% of all Dutch nursing homes providing care exclusively for somatic residents (somatic nursing homes; N = 44) and nursing homes with both somatic and psycho-geriatric wards (combined nursing homes; N = 242). Stratification was made by the type of nursing home (somatic versus combined) and by the size of the nursing home. For this latter, somatic nursing homes were divided into homes with 100 beds or less (N = 21) and homes with more than 100 beds (N = 23). Combined nursing homes, which are generally larger, were divided into homes with 200 beds or less (N = 142) and homes with more than 200 beds (N = 100). Two somatic nursing homes were randomly selected, one with less and one with more than 100 beds. Thirteen combined nursing homes were randomly selected, 6 with less, and 7 with more than 200 beds (table 1).

Secondly, from these 15 nursing homes a random and weighted sample of 600 residents was taken, representing 1 % of all 51.174 beds available in Dutch (somatic and combined) nursing homes 22. The residents were randomly selected by the researcher from a (anonymous) list of all residents. The number of selected residents per nursing home was weighted by the total number of beds destined for rehabilitation and long-term care in the involved nursing home. Only residents over 55 years of age were included, because the study was designed to focus on health-care for the elderly. Furthermore, residents had to be admitted to the nursing home for either rehabilitation services or for long-term institutional care. Patients consent was not needed according to the Dutch 'Regulations on medical research involving human subjects', since data were collected anonymously (using residents' numbers), and residents were not subjected to any treatment. As a consequence, there was no decline to participate.

Data from these residents were collected through interviews with the nursing home physician and the physiotherapist assigned to the resident. As part of these highly structured interviews, a questionnaire was filled out by the researcher for each resident, in the presence of the physician and therapist, enabling them to consult the medical file of the resident simultaneously. The domains in the questionnaire were based on the Minimum Data Set of the Residents Assessment Instrument (RAI)23 and determined in mutual agreement with a number of nursing home physicians and physiotherapists. The following areas of the RAI were addressed: communication and hearing patterns, physical functioning and structural problems, mood and behaviour patterns, disease diagnoses, oral and nutritional status, skin conditions, cognitive patterns, vision patterns, and continence. All information was based on the residents' medical file and on the judgement of the nursing home physician and the physiotherapist. No additional measurements were taken.

Two outcome, or dependent, variables were used. The first, dichotomous, outcome variable was whether the resident had received PT in the six months preceding the interview. All therapeutic intervention as registered in the medical file, was designated as treatment given, without time restrictions on the amount of treatment per week. The second, continuous, outcome variable was the amount of PT provided to the resident during the last 6 months, estimated by the PT in minutes per week.

The 'predictor', or independent variables were divided into two groups on the basis of their level, that is the characteristics on the level of the residents or characteristics on the level of the nursing homes (table 2). The medical diagnosis and co-morbidity, as determined by the nursing home physician, were taken into account on the level of residents' characteristics. The four medical diagnoses most frequently found among the 600 residents, stroke, dementia, status after total hip replacement, and non specified neurological disorders, were included in the analysis. Other diagnoses were treated as a reference group. When residents had more than one medical diagnosis, the diagnosis most determining the state of health, according to the nursing home physician, was used. Co-morbidity was defined as the number of medical diagnoses, excluding the main diagnosis. While 'co-morbidity in functioning' was defined as the number of (additional) impairments and limitations in activities.

Furthermore, we asked the nursing home physician which impairments or limitations had the most impact on the residents' health status. The impairments mentioned were categorised as 'impairments in mental functions' or as 'limitations in mobility and self-care'. Impairments and limitations in activity that could not be classified into one of these two categories, such as impairments in functions of the skin, genitourinary functions and limitations in communication, were treated as a reference group.

Additionally, limitations in communication, limitations in behaviour, age, sex, the length of stay since admission, the reason for admission (rehabilitation versus long term care) and whether the resident was staying on a somatic or psycho-geriatric ward, were used as independent variables 2425. On the level of the nursing home characteristics the volume of available PT, measured as full time equivalents (FTE) per bed with 1 FTE being 36 hours a week, the type of nursing home, the presence of a stroke unit and the degree of urbanisation were taken into account 2627.

Descriptive statistics for all variables were calculated with SPSS 11.5. Multilevel logistic regression analysis, carried out by MLwiN 1.1, was used to define the amount of variation in the provision of PT between nursing homes, and the factors that determine whether a resident receives PT or not. By means of multilevel linear regression, the amount of variation in the mean volume of PT between nursing homes and the factors that act on this volume were analysed. This latter analysis was carried out for the subgroup of residents that actually received PT. Multilevel analysis was used because the data had an intrinsic hierarchical nature. The sample of residents, level 1, is nested in the sample of nursing homes, level 2. The data, therefore, were not from independent observations, violating a major assumption of traditional regression analysis. In multilevel analysis, multiple levels are taken into account and in linear multilevel regression the proportion of variance at each level can be calculated. The order of adding predictor variables was determined by their level. The analysis was carried out in four steps:

1. Firstly, an intercept model was made, which is a model without any independent variables. This model estimates, in case of logistic regression, the mean percentage of the nursing home residents that receive allied health care. In case of a linear regression, the intercept model estimates the mean volume of therapy that is provided to the recipients. This latter model also establishes the contribution of both levels to the variation in the volume of treatment.

2. In the next step, all independent variables at the level of the residents were added. The contribution of each factor is expressed by the regression coefficient (β) and the standard error (SE).

3. In the third step, all independent variables on the level of the nursing home were added.

4. In the last step, only the factors that significantly contribute to the utilisation of PT are left in the model, that is the factors for which the quotient of β and SE is larger than |1.96|, the level of significance being 0.05.

For categorical independent variables indicator coding was used with the last category in each group treated as the reference group. The continuous independent variable residents' age was centred around the median age.

This study is the first one to examine how many Dutch nursing home residents receive PT and the extent to which the provision of PT varies across nursing homes. For this purpose, a random sample of nursing homes was taken, stratified for type and size of the nursing home, leading to representative representation of the population of Dutch nursing homes (table 1). The number of randomly selected residents was weighted by the total number of beds destined for rehabilitation and long-term care in the involved nursing homes. In addition, the proportion of residents living at a somatic ward and residents living at a psycho-geriatric ward in our sample, reflected the proportion of beds destined for somatic and psych-geriatric care in the Netherlands (table 1).

On average 69% of the nursing home residents received PT. In comparison with the international literature this is a rather high percentage of the nursing home population. This may, however, be partly due to different definitions of receiving treatment. In this study, all physiotherapeutic care as registered in the medical file, was designated as PT, without time restrictions for the amount of treatment per week. Adopting a minimum restriction of 30 minutes PT a week, as was used by Berg et al (1997) in some of the investigated countries, would result in 50% residents being treated 13. Furthermore, this high percentage of treatment recipients is probably due to the fact that physiotherapists are almost always employed by Dutch nursing homes, making referral easier. However, the percentage of residents receiving PT ranged significantly from 35% to 90%, which reflects a disparity in the care far beyond our expectations. Since we controlled for residents' characteristics, this difference can not be explained by a variation in the characteristics of the nursing home population as far as is measured in this study. In contrast, the availability of PT, varying from 1.12 to 3.33 FTE per 100 beds, contributes significantly to the difference in the percentage residents receiving treatment among nursing homes. Nursing homes, together with the nursing and rehabilitation facilities they provide, are mainly covered by the Exceptional Medical Expenses Act (AWBZ). In addition, residents contribute to the nursing home, depending on their personal financial resources or income. Inequalities in the financial resources of nursing homes may therefore explain some of the differences in the availability of physiotherapists, but we expect that divergent management policies of the nursing homes also play a role. Allied health care facilities will not always be given priority by management.

In this study the medical diagnosis was not associated with whether or not a resident receives PT. This finding is, however, somewhat biased by the fact that most residents with dementia stay on a psycho-geriatric ward, leading to a high correlation between the factor 'dementia' and the ward a residents is staying on. So when the factor 'somatic ward' is left out of the analysis, residents with dementia are less likely to receive PT, leaving all other predictor variables unchanged. Furthermore, residents admitted for rehabilitation and residents living on a somatic ward were more likely to receive PT than residents admitted for long term care or residents living at a psycho-geriatric ward, irrespective of their medical diagnoses. While it may be obvious that residents admitted for rehabilitation need physiotherapeutic care, it carries the risk that residents admitted for long-term institutional care do not receive the treatment they might need, because therapists assume that treatment is less effective for this group and residents admitted for rehabilitation have precedence in cases of limited availability of PT. However, the studies by Arlings and by Przybylski (1996) revealed that increasing the amount of PT and OT did have a positive effect on the functional status of long-term care residents 8. The pattern found in our study, that residents living at a psycho-geriatric ward receive PT less often (table 6), also reflects the belief that especially residents with good cognitive functioning are likely to benefit from treatment, whereas several studies demonstrate that geriatric patients with cognitive dysfunction show similar gains in functional status following specialised rehabilitation 42829.

So despite the fact that there is evidence that PT can be effective for long-term residents and for residents with low cognitive functioning, these residents receive less treatment. This suggests an under-use of PT for long-term residents and residents with low cognitive functioning. It is, therefore, recommended that physiotherapists working in nursing homes carefully reconsider which residents may benefit from treatment. This may need a shift in the focus of the physiotherapist, and other rehabilitation disciplines, from 'recovery and discharge' to 'quality of life and well-being' of residents. The relevance of such a change of philosophy is illustrated by the fact that a relatively large proportion of elderly residents who have been admitted for rehabilitation, will not be discharged home again and eventually remains at the nursing home receiving long-term care.

In our study, men were more likely to receive PT than women, irrespective of their medical condition. The reasons for this can only be hypothesised. Since most women survive their husbands, men may be more motivated for treatment and have more opportunities to return home, if their wife is capable of providing the necessary care and social support 30. However, the actual reasons remain unclear and the little literature available on gender differences regarding PT in nursing homes reveal conflicting findings. Whereas Berg et al (1997) found that men were less likely to receive rehabilitation services in Italian nursing homes, Yeh (2004) and Kupper (2006) found significantly more men among physiotherapy recipients of PT than could be expected by chance alone. Obviously, further research is needed to explain these gender differences.

This study has some limitations. It was not possible to take the severity of the residents' medical problems and impairments into account, due to the method of gathering data. Data were collected through interviews with the nursing home physician and the physiotherapist, and were based on the residents' medical file. The severity of the patient's condition or their functional status were not asked, since these aspects are not always registered in a standardised manner, making a comparison between different nursing homes difficult. There are, however, no reasons to expect systematic differences in the severity of the medical problems between the residents populations of the 15 nursing homes. Another limitation is that the accuracy of assessing the medical diagnosis and the presence of impairments in mental functions and limitations in mobility and self-care, relies on the nursing home physician reports. Since diagnostic procedures used for establishing the residents' health problems were not standardised, they may differ between nursing homes.