Systematic Review of the Use of Hospital Administrative Data to Assess Functional Decline

,


Introduction
One of the biggest challenges faced by health policy makers is to prevent functional decline in the elderly population [1]. This population is increasing and it is estimated that it will double by the year 2050 [2]. Functional decline, frailty and disability increase with age [3]. It is foreseen that the burden of functional decline and disability will markedly increase soon. In the US, the National Committee on Vital and Health Statistics acknowledged the importance of functional health. It stated that 'achieving optimal health and wellbeing for Americans requires an understanding across life span of the effects of people's health conditions on their ability to do basic activities and participate in life situations -in other words, their functional status' [4].
It is important to predict patients with functional decline because they are a higher risk of the use of healthcare resources5. Previous prediction models for functional decline have obtained data from questionnaires and surveys, which ask questions about activities of daily living (ADL), such as cooking, bathing, toileting and climbing stairs [5]. Questionnaire-based studies are expensive and difficult to conduct at a population level as everyone must undergo detailed assessment of their functional health. Proxy or self-reporting methods used in these studies rely on patient self-assessment for data collection which cannot be verified objectively. A significant number of these patients suffer from cognitive and mental health disorders, making it difficult to retrieve information for questionnaires. Moreover, data gathered by conducting surveys provides cross-sectional overview with limited follow up.
Administrative data can provide an effective alternative and has the potential to evaluate functional decline [5][6][7][8][9]. Administrative data offers 'system wide information about health conditions and services in a consistently coded format' [4]. It is primarily collected for billing purposes but it can be used for research [5]. It is collected for the entire population, and can be tracked over a long period. The data can be linked to other data sources that can provide additional information, such as patient's demographics, medical history and pharmacotherapy. The data collection is not biased by recall of patients as in self-reported surveys [6]. Systematic diagnostic codes are used to record comorbidities. There are various outcome metrics that are available, for instance, recurrent hospitalization, increased length of stay (LOS), and discharge to nursing home. The aim of the study is to find out whether administrative data can be used to predict functional decline in the patients by conducting a systematic review of the literature.

Search strategy
The methodology of the systematic review was based on PRISMA guidelines ( Figure 1) [10]. The PICOS process was used to develop the search strategy [10]. to identify studies involving functional decline and hospital administrative data (Table 1). Boolean terms, like 'OR' and ' AND' , were used to combine search terms.
Further studies were identified through cross-referencing of initial studies reviewed. Two independent researchers, AS and AR, reviewed the selected studies separately. Meta-analysis was not conducted because there was significant heterogeneity in the studies.

Inclusion and exclusion criteria
The following inclusion criteria were used: 1. Participants: adult patient population over the age of 18 diagnosed with any type medical condition.

Intervention:
The patient admitted to hospital for any medical or surgical condition and underwent any intervention to prevent or reduce functional decline.
3. Comparison was made between studies using HAD and any other method of evaluation of functional decline.

4.
Outcome: assessment of functional decline following hospital admission.

5.
Study design: studies that used HAD based metrics to derive or validate model to assess functional decline.
The following exclusion criteria were used: 1. Studies that did not use HAD were excluded from the review.
2. Studies that used clinical data from controlled trials, observational studies, case series or clinical registries.
3. Studies only evaluating cost outcomes.

Data collection
Basic demographics were obtained from each study included in the review. Year of publication, place of data collection, administrative databases used, and aim and objectives of each study were recorded. Information on methodology of each study was collected, such as, number of patients, use of control group, diagnosis of patients, types of outcome measures used, results of significant outcomes, and follow up period.

Assessment of risk of bias
The Newcastle-Ottawa scale was used to assess bias in the included studies [11]. This scale is validated and recommended by Cochrane review methodological guidelines for non-randomized studies. Criteria for assessing bias associated with the study was based on selection of patients, comparability of methodologies or intervention and clear definition of outcomes measured. A study can get a maximum of 8/8 stars, suggesting minimum bias.

Results
There was a total of 4 studies included in the systematic review ( Table 2). Four studies evaluated functional decline in elderly population [12][13][14], while the other two studies included adult population over the age of 16 [15,16]. The patient population selection differed in studies; it included community dwelling elderly population [13], patients admitted to inpatient rehabilitative facilities [14,16], patients who underwent spine surgery [15], and stroke patients [14].
The hospital administrative data used by the studies were local Canadian [15] and American [12] hospital administrative databases, and Medicare database [7,13]. The information on ADL for comparison with HAD-based variables was obtained from survey data [7,13] and proxy or self-reported questionnaires [12,15]. We identified three predictive models that used HAD to assess functional decline.

Hospital readmissions
Coleman et al. aimed to compare predictive accuracy of two indices for functional decline, one based on hospital administrative data and the other on self-reported questionnaire [12]. There were 1,764 patients included in the study and they were followed up for 4 years. The previous number of readmissions were used as an outcome measure from HAD to predict functional decline. Self-reported functional status was measured using SF-36 questionnaire. The outcome measure was prediction of restricted activity days (RAD), which was derived from cumulative length of stay in hospital for patients for every following year. The predictive accuracy for the model based on readmissions obtained from the administrative variables was not significantly different from self-reported model for functional health. The Area Under the Curve (AUC) depicting predictive accuracy of administrative model was 0.691 as compared to 0.714 for self-reported model (P=0.144).

Procedural claims-based predictors
Davidoff et al. and Faurot et al. created a predictive model based on different types of procedures and medical equipment that patients undergo and use. The studies were conducted on the elderly patient population over the age of 65 [7,13]. The studies extracted information from US-based administrative data, CMS (Centers for Medicare and Medicaid Services). The data includes the type of services provided and procedures conducted on each patient. The procedures were coded in various formats: ICD-9, the American association Current Procedural Terminology (CPT) and CMS Healthcare Common Procedure Coding System (HCPCS). The outcome measure was self-reported information on decline in ADL and mobility. It was obtained from CMS data linked to Medicare Current Beneficiary Survey (MCBS) and was used to validate the predictive model.

Post-operative outcomes
Omoto et al. assessed the use of PCSI (post-operative crosssectional imaging) and reoperation as predictive indicators for functional decline [15]. The use of PCSI included either magnetic resonance imaging (MRI) or computed tomography with myelography (CT-myelogram). It was proposed that the use of PCSI and reoperation were associated with functional decline in patients who underwent spinal surgeries (discectomy, decompression (laminectomy/ laminotomy) or fusion for disc herniation). The model was validated by comparing it to questionnaire-based functional outcome measures (SF-36 and Oswestry disability index). There were 148 patients included in the study and the patients were followed up for 2 years. There was no significant relationship between the occurrence of PCSI or reoperation and functional decline.

Summary of results
The review was based on 6 studies that developed 3 predictive models based on HAD quality metrics to assess functional health. Hospital readmission and specific procedural codes based on HAD were used as a proxy to measure functional health. They showed significant correlation with functional decline. The model based on PCSI and reoperation did not show any significant association with functional health. The study population, type of administrative data, and outcome measures differed in the studies included in the review. Similarly, studies in the review had variable ranking based on the Newcastle-Ottawa scale for assessment of bias. Common reasons for high risk of bias were inadequate follow-up and not clearly defined comparison group [7,15]. Two models assessed functional health but used this as a proxy to measure performance status and frailty [13].

Comparison with previous studies
Previous studies have shown strong link between hospital readmissions and poor quality of life [12,16]. Patients with higher number of hospital readmissions had prolonged length of stay in the hospital every year and suffered higher morbidity [14]. They were also associated with increased risk of discharge to nursing home [11]. Hospital readmissions directly impact functional status of the patient as shown in the previous studies where annual survey was conducted at the rehabilitative hospitals [12]. Each hospital admission cause loss in muscle and bone density, put one at an increased risk of dehydration and malnutrition, and risk of iatrogenic injuries. All these conditions can lead to patient having falls, fractures and delirium. These patients lose the ability to cope at home and require further readmissions [12].
In addition, hospital readmission is one of the most commonly recorded HAD based metric. This information is commonly extracted and used in different studies to assess patient outcome and used as a marker of poor health status [17][18][19]. Similarly, our review showed that the predictive model based on hospital readmission had better sensitivity and specify than the other models. The accuracy to predict functional decline for model based on hospital readmissions was like self-reported questionnaire.
Predictive models based on FIM score and FIS staging had limited applicability for National and International comparison of functional outcome as observed in previous studies [19]. In general hospitals, there are no regulations to record ADL [14]. Its current use is limited to specific institutions and regions because the recording of FIS requires a significant time investment by trained nursing staff at the time of admission and discharge of the patient [15]. The recording of ADL is conducted in hospitals based on financial incentives give to them. Similarly, in our review, FIS score was derived from information on ADL collected by administrative databases of rehabilitative hospitals [14]. It is still to be investigated how these scores are recorded in busy general hospitals where turnover is fast and patients are moved from one unit to another.

Strengths of the review
This review attempted to combine limited data available on the measurement of functional health using HAD based outcomes. Certain models were identified that could potentially be used to study functional health. Various databases were searched to identify studies that could be included in the review. Newcastle-Ottawa scale was used to measure bias associated with the studies. This review suggests that further research is required to completely assess potential of administrative data to evaluate functional decline. The combination of outcomes, such as length of stay, discharge destination and causespecific readmission rate, may provide better predictive ability to assess functional decline. Although discharge destination, LOS and mortality were other outcome measures based on HAD that had been used in previous clinical studies to assess impact of various treatments on patient's health but they had not been assessed for their direct association with functional health [7,12,14,16]. Once it can be shown that outcomes based on HAD can be used as a validated proxy measure for functional health, its application can be implemented at a larger scale as in most Western countries hospital administrative data is routinely and annually collected. It can be used to study trends in the changes in functional health at a population level and help assess clinical factors and interventions that can prevent functional decline.

Limitations of the review
There were limited studies that assessed the role of administrative data to study functional decline, hence, the number of studies included in the review is small. There was significant heterogeneity among the included studies. They had different outcomes measures, tools to validate predictive model, and patient cohorts. It was not possible to combine the results and perform meta-analysis of the outcome. The studies that used claims-based codes as a marker for functional decline has limited generalizability [13]. It is still uncertain how often the information on health care service utilization and procedural codes is recorded in different regions. Coding of certain procedures is not widespread and vary in different databases [13].
The use of coding for post-operative cross-sectional imaging (PCSI) and reoperation as a proxy measure of poor functional health in lumbar spine patients was investigated but it was not found have any significant correlation [15]. The authors concluded that reoperation rate was a poor marker as significant numbers of lumbar spine patients undergo a planned second operation. These metrics therefore have limited use as a proxy measure for functional health.

Conclusion
Three predictive models have been developed to assess functional decline based on outcomes derived from hospital administrative data. Models based on hospital readmissions have the potential for widespread use because it had significant correlation with functional decline. Its predictive accuracy was like self-reported functional health. However, further studies are required to completely assess potential of administrative data to evaluate functional decline.