The Resident Assessment Instrument (RAI), developed in response to 1987 US Nursing Home Reform laws, aims to provide a comprehensive assessment of nursing home residents . The cornerstone of the RAI is the MDS, a uniform, standardized, computerized tool for assessing multiple domains of a person's physical, social and psychological health and function, including skin integrity, number and stage of pressure ulcer(s) and typical risk factors for pressure ulcer development noted in the literature . The MDS assessment tool has received extensive reliability and validity testing . The RAI User's Manual directs nursing home staff to complete a full MDS evaluation of each resident by admission day 14, annually, and upon major change in functioning, with a subset of items completed quarterly . The RAI also includes eighteen RAPs that are triggered by specific MDS item responses. RAPs provide clinicians with support for conducting a more in-depth assessment for prevention and care planning of potential or actual problems, such as Pressure Ulcers . These RAPs have recently been updated as part of an international effort to combine new knowledge gained from the use of the RAI around the globe with current best practice guidelines in a variety of areas including pressure ulcers .
RAI Version 2.0 has been utilized in the US since 1996, and Ontario LTC homes began implementing it as the standard of care in 2005. It has been translated into 18 languages and is used internationally for care planning, facility management, needs assessment, policy development, quality improvement and benchmarking, reimbursement, research, or service eligibility, making it the most widely used comprehensive assessment instrument world-wide. An updated version has recently been completed by the interRAI research network (http://www.interrai.org) .
The Braden Scale for Predicting Pressure Sore Risk  is likely the most widely used tool in facility-wide pressure ulcer prevention programs in North America. The Braden Scale is a summary measure comprised of six sub-scales for measuring an individual's functional determinants of pressure and tolerance of tissues to withstand pressure. Each of the six subscales (activity, mobility, sensory perception, nutrition, moisture, and friction and shear) is scored 1 (least favourable) to 4 (most favourable), with the exception of friction and shear which is scored from 1 to 3. Scores can range from 6 to 23; lower scores are associated with higher risk for developing pressure ulcers.
Risk assessment tools, including the Braden Scale, have been criticized for their generally weak properties , although among tools with published findings, a recent review article found the Braden Scale to perform better than others . The Braden Scale domains of nutrition and sensory perception have been shown to have poor reliability . It is notable that there is a lack of good evidence that the use of any pressure ulcer risk scale actually reduces pressure ulcers in clinical practice .
The goal of this work was to investigate the possibility of an MDS-informed pressure ulcer risk scale that could eliminate duplicated assessment burden. Being the product of three parties (wound care education, government health care administration, health research), the investigation was often formative in nature, characterized by several phases:
Identification of existing LTC homes (a small convenience sample) where both Braden Scale and MDS data were collected concurrently, and using wound care expertise along with evidence from these data to map the Braden Scale using MDS items,
Testing the ability of this mapped Braden Scale to predict a new pressure ulcer among those without a pressure ulcer, in a much larger sample,
Consideration of other MDS items that might improve predictive performance.
Setting and Subjects
Beginning in 2005, Ontario LTC homes volunteered to implement the MDS assessment as standard practice, with groups of homes receiving training and support in a phased roll-out to all homes in the province. By the fall of 2007, 89 of the province's more than 600 homes had submitted MDS data to the national reporting system at the Canadian Institute for Health Information (CIHI). Data included an admission or baseline MDS assessment of each resident having a stay of 14 days or longer, and subsequent reassessments every 90 days (or earlier in the case of a clinically significant change in health or functional status). A total of 72,602 assessments, with a resident identifier allowing longitudinal linkage but keeping identity anonymous, were available for analysis. From these data, an analytic dataset was created of all unique individuals using their baseline assessment/reassessment pair, for a total of 14,083, of which 91.6% (n = 12,896) had no recorded pressure ulcer at the baseline assessment. The median time between assessments was 91 days.
Three LTC homes were identified as collecting Braden Scale scores and MDS assessment data. The Braden Scale information included the 6 sub-scores and were captured electronically and linked through a non-real-world identifier to extracted MDS data, yielding a dataset of 257 cases. The median time difference between MDS and Braden Scale assessments was 9 days.
In addition to the derivation data from Ontario LTC homes, data from other settings were drawn from holdings available to interRAI researchers, in order to test pressure ulcer prediction. Ontario CCC hospital MDS assessments were drawn from 2002 to 2007. This population differs from Ontario LTC home residents, primarily in that they tend to have shorter stays, require post-acute care, are more clinically unstable, and are more likely to receive rehabilitation services. A baseline assessment/reassessment sample was constructed using the same methodology as for LTC homes, resulting in a sample of 17,956, of which 72.7% (n = 13,062) had no recorded pressure ulcer at baseline. Also of interest are long-stay home care clients assessed with the RAI-Home Care (RAI-HC)  by Ontario home care case managers as part of routine clinical practice. A sample of community-assessed clients was drawn using a similar baseline assessment/reassessment approach, resulting in 76,068 clients, of which 96.2% (n = 73,183) had no recorded pressure ulcer at baseline. A subset of these home care clients assessed in acute care hospital and identified for long-term care placement was identified for comparison. In addition, a cross-sectional research sample assessed using the interRAI Palliative Care instrument  was available (n = 988), to explore the characteristics of community-dwelling palliative clients and pressure ulcers in Ontario.
MDS Assessment of Pressure Ulcers
Assessors trained in the MDS record the presence, stage, and number of pressure ulcers in the last 7 days. Ulcers were staged by the following criteria: (1) persistent area of redness (without a break in the skin) that does not disappear when pressure is relieved, (2) a partial thickness loss of skin layers that presents clinically as an abrasion, blister or shallow crater, (3) a full thickness of skin is lost, exposing subcutaneous tissues (presents as a deep crater with or without undermining adjacent tissue), and (4) a full thickness of skin and subcutaneous tissue is lost, exposing muscle or bone .
For the matched Braden Scale-MDS dataset (n = 257), Spearman rank correlations between each of the 6 Braden sub-scores and candidate MDS items and scales (identified in a series of telephone meetings among the three parties) were examined in order to inform a possible cross-walk of the Braden Scale to the MDS. Subsequently, various constructions of a Braden Scale cross-walk were considered, informed both by clinical insight/face validity as well as strength of correlation. The cross-walk scale versions were constructed to mimic the Braden Scale both in sub-scale and total score ranges. Insufficient pressure ulcer incidence data were available for this small dataset, so these cross-walk algorithms were applied to the larger baseline/reassessment data in LTC homes to examine performance in predicting new pressure ulcers among those without pressure ulcer at baseline. The cross-walk versions were evaluated using logistic regression, predicting a new pressure ulcer at the next quarterly assessment, with the C-statistic (area under the receiver operator characteristic curve) as the main evaluation criterion. The interpretation of the C-statistic of 0.7, for example, in this case is that a randomly chosen individual who goes on to develop a pressure ulcer is likely to have a higher scale score than a randomly chosen individual who does not develop a pressure ulcer 70% of the time. Values of 0.5 reflect no better than chance alone, and higher numbers indicate better diagnostic prediction, with a value of 1.0 being perfectly accurate . The value of the C-statistic will be used to differentiate relative performance of differently constructed scales within the same data.
Keeping the Braden Scale findings in mind, a series of exploratory analyses was subsequently done using the larger Ontario LTC homes baseline/reassessment dataset alone, to explore if other constructions of MDS items could improve on the Braden Scale cross-walk in predicting new pressure ulcers. Here the methods included both multivariable logistic regression (to test independent effects) and decision tree modelling (to try to discover potential interaction effects), using as the dependent variable new pressure ulcer at reassessment among those without a pressure ulcer at baseline. Evidence from the literature and clinical expertise contributed to the refinement of these models, with the goal of a predictive scale that could be calculated using an existing MDS assessment, but also could be readily done at the bedside prior to completion of the full set of MDS items by admission day 14, or prior to a scheduled MDS reassessment. Logistic regression of new pressure ulcer was used to determine those MDS items statistically significant (at p = 0.05 or below) in a multivariable, parsimonious model. These items were applied in an interactive decision tree model (with potential splits based on chi-square values for a new pressure ulcer and the candidate splitting variable) as a final step to see if alternative treatment of them might improve predictive performance.
The final scale was replicated in other interRAI instruments, here the home care and palliative care tools to examine how it was associated with incidence of pressure ulcers in other at-risk populations.
SAS and SAS Enterprise Miner Tree Desktop version 9.1.3 were used for all analyses.
Data sharing agreements allowed for the transfer of anonymized data from both CIHI and the Ontario Ministry of Health and Long-Term Care to the University of Waterloo, and ethics approval for analysis of anonymized data was in place from the Office of Research Ethics, University of Waterloo.