Development of a Risk Prediction tool for entering a Nursing Home in those aged 65 and over by Scottish Collaboration for Public Health Research & Policy (SCPHRP)

Chief Scientist Office Final report Form

Form 4

SCPHRP reference number: 0309/10

Please complete this form in Verdana 10 point font size Project title: Development of a Risk Prediction tool for entering a Nursing Home in those aged 65 and over in a Scottish Population

Start date:1/3/2010

Finish date:28/02/2011

Investigators: Prof Peter T. Donnan

Dr Pauline Lockhart

Prof Frank Sullivan

Dr Wendy Gidman

Dr Colin McCowan

Prof Bruce Guthrie

Dr Michael Norbury 1.

Summary

Background. Entering a nursing home in later life is associated with enormous impact on the health of an individual, as well as cost consequences for health services. A first step before effective interventions to maintain autonomy can be developed is to have the ability to identify those who are at greatest risk. A number of mainly US studies have identified illness severity, being female, lack of caregiver support, cognitive disability and previous hospitalisations as key predictors of entering a nursing home. In contrast, there has been little work on predictors in Scotland or the UK where priorities and organisation of health care differ. Linkage of relevant data in the Health Informatics Centre in Tayside will facilitate the development of a predictive algorithm. Having the ability to predict in advance, those at high risk could lead to early interventions to allow patients to remain in their homes longer with benefit to both patient and the NHS. Methods. This record-linkage study of Tayside population health databases assessed the risk of entering a Care Home in the whole population aged 65 or over in the period 2005 â&#x20AC;&#x201C; 2009. A baseline was ascertained for each individual in the cohort who had never been in a Care Home and their history over the previous year of prescriptions received, hospitalisations, deprivation decile based on last community residence, and date of death were extracted. The binary outcome was defined as entry to either a Nursing Home, or Mixed Nursing Home and residential care, or unknown type coded as entry to a â&#x20AC;&#x2DC;Care Homeâ&#x20AC;&#x2122; (Yes, No). Follow-up to entry to a Care Home or not was analysed with logistic regression and Cox regression models and predictive models derived. The discriminative performance of the models were assessed by calculating the c-statistic (AUROC for logistic) and calibration assessed with the Hosmer-Lemeshow test. Results A total of 84,432 people aged 65 or over were identified in the cohort who had never entered a Care Home. The rate of entry was 5.6% over a median follow-up of 5 years. Strong predictors of entry were older age as a quadratic function, with an

increasing rate for older women relative to men. The affluent were most likely to enter a Care Home and the difference between affluent and deprived was greatest in women. A high number of days in hospital in the previous year were associated with higher risk as was previous admissions for cerebrovascular disease, falls, and diabetes. Receipt of drugs for dementia, antipsychotics, and drugs for Parkinsonâ&#x20AC;&#x2122;s disease as well as the total number of prescriptions were all independently strongly associated with entry to a Care Home. A number of medications were associated with lower risk; namely, receipt of NSAIDS, statins, ulcer-healing drugs and beta-blockers. A final model with logistic regression was derived that included 20 factors which gave an AUROC = 0.89 and good calibration (p = 0.210). A similar model was derived from the Cox regression but with further factors added as power was greater with full follow-up utilised. Conclusions This population study demonstrated that linking demography, community prescribing and hospitalisation datasets allowed powerful predictive algorithms for entry to a Care Home to be derived. The models need to be externally validated and developed into user-friendly software to facilitate evaluation in the population in terms of patient and NHS outcomes. 2.

Original aims

The aims of the study were to examine risk factors associated with entry to a Nursing Home in those aged 65 or over in Tayside, Scotland, derive a risk equation for entry to a Nursing Home and then test its predictive ability and accuracy and finally, to create a simplified scoring rule for entry to a Nursing Home for clinical use.

Research questions a) What risk factors are associated with future entry to a Nursing Home and which of these are most important? b) Can these factors be combined into a statistical model that predicts future entry to a Nursing Home? c) Does the derived model have adequate predictive ability and does it accurately predict entry to a Nursing Home using internal validation? d) Can a simple prediction rule be derived for clinical use?

Methodology

Study population This consisted of patients alive and living in Tayside at some point during 2005 to 2008. Patients had to be 65 years old at follow up, therefore only patients aged 65 years on 01/01/2005 or who had their 65th birthday before 31/12/2008 were included. Patients who enter Tayside during the study period are followed up at least two years from their date of entry to allow one full yearâ&#x20AC;&#x2122;s history (Medication Exposure) to be obtained followed by at least one year follow-up. Data Sources The Health Informatics Centre (HIC) residing at the University of Dundee holds data for the population of Tayside (~400,000 inhabitants) which is reasonably representative of the UK population. Since the 1980s, patients have been tracked within the NHS, the sole health provider, with a Community Health Index (CHI) number. This number allows tracking of patients for drug use, laboratory testing, hospitalization, GP visits, mortality and other data. Tayside data (http://www.dundee.ac.uk/hic/data/tayside/) also contain disease-specific datasets for diabetes, respiratory and thyroid disease. Various datasets can be linked and anonymised by HIC researchers and analyses can be undertaken, provided research proposals are registered and approved, and executed locally[1]. All data linked in HIC is anonymised under strict SOPs with Caldicott Guardian approval and annual external audit.

Cohort definition Data for those aged 65 and over were extracted for the period of 2005 – 2008 from the population of Tayside by HIC, along with the community Health Index (CHI) to facilitate deterministic linkage. A baseline was ascertained for each individual and their history over the previous year of prescriptions received, hospitalisations, deprivation decile based on last community residence, and date of death were extracted. Those initially identified at baseline as being already resident in a Nursing Home were excluded. This cohort was then be followed-up for at least one year to the end of the study, date of death or date of entry to a Nursing Home, whichever was the earlier. Definition of baseline Study Start Date: 01/01/2005 Entry Date: Patients require one full year history prior to start of follow up. • • •

One year prior to Start Date (if 65 and lived in Tayside for more than a year) One year prior to date of 65th Birthday (unless patients have not been living in Tayside for one full year prior to 65th birthday. For these patients Entry Date equals the date they entered Tayside). Date Entered Tayside (If 65 or older or turn 65 in first year).

Baseline Date (Start of Follow Up): Entry Date plus one year. • Study Start Date (if 65 and have lived in Tayside for more than one year) • Date of 65th birthday (unless entered Tayside less than one year before). • One year after date entered Tayside (If 65 or older or turn 65 in first year). Exit Date: End of patient follow up, the earliest of the following three dates: • • •

Study End Date (if alive and still residing in Tayside) Date moved out of Tayside Date of Death

Study End Date: 31/12/2009 Potential Predictors Demography: Age, Gender and decile of Scottish Index of Multiple Deprivation. Previous hospitalisations: total hospital days in previous year, number of previous hospitalisations, reason for last admission based on ICD code in categories such as falls (ICD10 W00-W19), CHD (I20-I25, I44-I50), Cerebrovascular Disease (I61, I63-I69, G45), CHF (I50), Cancer (C00-C99), diabetes (E10/11/12/13/14), respiratory conditions (J00-J99), length of stay of most recent admission and time from most recent admission to Baseline Date (days). Previous medication use in last year: based on BNF categories such as antidepressants, anti-psychotics, anticoagulants, NSAIDS, antiplatelet agents, diuretics; etc[2]. These are fully listed in Appendix 1. Social support: Living alone status (Yes, No) was to be identified from the known address and the number of people living at that address, at that CHI snapshot. This was obtained over time with annual CHI snapshots of the whole population. In addition to the above factors, we considered those factors used in predicting Emergency Admissions in the Scottish Patients At Risk of Readmission and Admission (SPARRA) and Predicting Emergency admissions Over the Next Year (PEONY) [3] models.

Outcome assessment The Definition of ‘In Care Home’ People aged 65 or older were identified with multiple addresses over the time period 1st Jan 2005 to 31st Dec 2009 (or aged >=62 at 1st Jan 2005 and were resident in Tayside at some point during that time). Nursing home or ‘In Care’ residence was identified from the address lists in Tayside and the number of people living at that address held by HIC. This was also checked against the Scottish Care Commission Register held in, Edinburgh (http://www.carecommission.com/index.php?option=com_content&task=view&id=16&Itemid=112). [4] The binary ‘In Care’ indicator included the following categories: 1) Nursing Home 2) Mixed Nursing Home and residential care 3) Nursing home of unknown type. Since category 2 is the largest it is more accurate to describe their accommodation status as ‘In Care Home’ rather than just ‘nursing home’ and so the term ‘Care Home’ is used in this study. In addition, Hospitalisations were obtained from the Scottish Morbidity Record (SMR01), the Tayside portion of which is held in the Health Informatics Centre [1]. Power and Sample Size In pilot work there were 70,299 patients aged 65-99 in Tayside in 2005 with 4,557 resident in care homes. Over the period 2005-2006 there were a total of 2,352 new admissions to a care home giving an event rate of approximately 3.6%. With this size of population it would be possible to detect Odds ratios > 1.1 with 80% power. In a multiple regression model as a rule of thumb a logistic model requires 10 events per variable considered. With a likely total of approximately 4,000 admissions there is more than enough power to consider for example 30-40 potential predictor variables. Statistical Methods Descriptive information on the characteristics of the baseline population were tabulated and summarised as means and standard deviations for continuous factors and as percentages and number for categorical data. Statistical modelling utilised both logistic regression for a fixed one year follow-up as well as time to event modelling such as Cox Proportional hazards models that use the full follow-up time. Since mortality will be high in such an elderly population, it may be necessary to consider competing risk models to obtain true risks of events. Selection of risk factors will utilise Akaike’s Information Criterion (AIC) which has two functions: 1) to prevent selection of spurious factors when sample size is large and hence reduces type 1 error; 2) is asymptotically equivalent to cross-validation and so does not require wastage of data by creating test and derivation datasets . Model performance will be assessed for predictive ability by estimating the cstatistic, as well as assessment of calibration or accuracy using a Hosmer-Lemeshow test[5]. Once the final model is derived, this will be simplified to give scoring rules that can be implemented in a clinical situation or at health board level [6] All analyses were implemented in SPSS (v 18) and SAS (v 9.2) statistical software. 4.

Results

The results are presented for each study question in turn below. 1. What risk factors are associated with future entry to a Care Home and which of these are most important?

Figure 1 below shows the derivation of the study cohort with exclusion of those who had entered a Care Home prior to the baseline.

Fig. 1 Derivation of the final cohort for analysis ( n = 84,432)

The characteristics of those in the cohort are presented in Table 1.

Table 1 Factors associated with future entry to a Care Home

Cohort

Characteristic

Mean age at baseline (range)

84,432

Care Home Admission Yes N (%) No N (%) 4717 (5.6) 79715 (94.4)

73.1 (65-110)

Gender

Men Women

SIMD deciles

1 least deprived 2 3 4 5 6 7 8 9 10 most deprived Missing

(43.8%) (56.2%) (6.9%) (8.1%) (9.2%) (10.4%) (9.7%) (10.4%) (11.1%) (11.1%) (11.5%) (10.7%) (0.9%)

Hospital usage in year before Baseline Date Median number of hospital admissions (days) Median number of hospital days (days) Median time since last admission over previous year (days)* Median LOS of previous hospital admission (days) Mean (SD) log transformed LOS Previous admission for:CHD Cerebrovascular Disease CHF Cancer Diabetes Respiratory Conditions Falls Medication use in previous year: Median (Range) No. of Scripts Ulcer healing drugs (BNF 1.3) Diuretics (BNF 2.2) Beta-blockers (BNF 2.4) Anti-hypertensives (BNF 2.5) Nitrates (BNF 2.6) Anticoagulants (BNF 2.8) Antiplatelet agents (BNF 2.9) Antifibrinolytic drugs (BNF 2.11) Lipid regulating drugs (BNF chap 2.12) Respiratory drugs (BNF 3)

81.6

72.6

1451 (3.9) 3266 (6.9)

35531 (96.1) 44184 (93.1)

153 (2.6) 361 (5.3) 145 (1.9) 273 (3.1) 534 (6.5) 468 (5.3) 597 (6.4) 747 (8) 667 (6.9) 594 (6.6) 178 (22.9) Median (Range)

5667 6468 7633 8539 7698 8289 8738 8628 9051 8406 598

(97.4) (94.7) (98.1) (96.9) (93.5) (94.7) (93.6) (92.0) (93.1) (93.4) (77.1)

0 (0-16) 0 (0-339) 366 (1-366)

Median (Range) 0 (0-46) 0 (0-323) 366 (1-366)

0 (0-271) 0.587 (1.220)

0 (0-418) 0.204 (0.659)

85 74 17 55 2 80 164

(1.8) (1.6) (0.4) (1.2) (0.0) (1.7) (3.5)

31 (0-657) 1373 (29.1) 2256 (47.8) 1021 (21.6) 1283 (27.2) 1544 (32.7) 243 (5.2) 2010 (42.6) 1 (0.0) 1083 (23.0) 596 (12.6)

1272 321 173 1042 37 657 637

(1.6) (0.4) (0.2) (1.3) (0.0) (0.8) (0.8)

20 (0-635) 21550 (27.0) 27477 (34.5) 18177 (22.8) 21962 (27.6) 23004 (28.9) 3612 (4.5) 23693 (29.7) 23 (0.0) 22642 (28.4) 10932 (13.7)

Hypnotics and anxiolytics (BNF 4.1) Antipsychotics (BNF 4.2) Antidepressants (BNF 4.3) Analgesics (BNF 4.7) Antiepileptics (BNF 4.8) Anti-Parkinsonian (BNF 4.9) Drugs used in substance dependence (BNF 4.10) Anticholinesterases (BNF 4.11) Antibacterial drugs (BNF 5.1) Diabetes (BNF 6.1) Osteoporosis (BNF 6.6) Anaemia (BNF 9.1) NSAIDS (BNF 10.1.1) *If no admissions in the previous year the value was 366

1035 (21.9) 307 (6.5) 1007 (21.3) 2343 (49.7) 193 (4.1) 162 (3.4) 40 (0.8) 201 (4.3) 1758 (37.3) 386 (8.2) 361 (7.7) 608 (12.9) 874 (18.5)

9404 (11.8) 999 (1.3) 9879 (12.4) 28959 (36.3) 2368 (3.0) 727 (0.9) 1378 (1.7) 187 (0.2) 27045 (33.9) 6167 (7.7) 3298 (4.1) 4369 (5.5) 16090 (20.2)

Overall the admission rate in this cohort of people aged 65 or over was 5.6% (table 1). As would be expected those who entered a Care Home were older and more likely to be women. For previous admissions to hospital in the year prior to baseline an admission for a fall or cerebrovascular disease were associated with higher risk of entering a Care Home. Note that unlike other reasons for admissions all fields were used in SMR01 for falls as the ICD code was found to be never the ‘main’ cause of the admission. The continuous measures of prior hospital use such as number of hospital days in the previous year were all associated with Care Home entry but as highly skewed the medians were all zero in table 1. Of the medications based on BNF chapters, drugs used for dementia (BNF 4.11), Parkinson’s Disease, diuretics, antidepressants, antipsychotics, hypnotics and anxiolytics, analgesics and drugs for anaemia were all strongly associated with entry to a Care Home. Interestingly, drugs such as lipid regulating drugs (Statins mainly), beta-blockers, NSAIDS and respiratory drugs were associated with lower risk of entry. One possible measure of ‘importance’ is the size of the association. Using a follow-up period of one year the Odds Ratios for all the above factors was estimated using logistic regression. Note that this reduced the number of events to n = 968 as only one year of follow-up was utilised. The results are shown in appendix 2. In terms of binary variables the order from highest OR first was: • Medication for dementia (BNF 4.11) OR = 13.7 (95% CI 10.1, 18.5) • Antipsychotics (BNF 4.2) OR = 8.7 (95% CI 7.1, 10.7) • Hosp admission in previous year for Cerebrovascular disease OR = 7.8 (95% CI 5.4, 9.3) • Hosp admission for falls OR = 7.1 (95% CI 5.4, 9.3) • Drugs used in Parkinson’s Disease OR = 4.9 (95% CI 3.6, 6.6) • Hosp admission for diabetes OR = 4.7 (95% CI 1.1, 19.4) Note that the Odds ratios are slightly misleading for continuous measures as these depend on the units used which by default is one unit, for example, one year for age. A better measure is the Wald statistic and for this metric the highest were: • Age • Number of days in hospital during last admission • Time from last admission to baseline up to maximum of one year • Number of prescriptions in previous year

Of course, all of these associations are unadjusted for confounders so a more thorough assessment using multiple regression methods is required. 2. Can these factors be combined into a statistical model that predicts future entry to a Nursing Home? The results of the multiple logistic regression modelling are shown in table 2. Note that as logistic regression models events over a fixed time period, in this case one year, then the number of events is smaller than when the full follow-up is used. One year follow-up results resulted in 968 events giving an annual rate of 1.1% for the whole cohort. Age was a strong risk factor and in the form of a quadratic which means that the odds of entry to a Care Home accelerated with age rather than just in a linear fashion (fig. 2). There were also a number of interactions with age, in particular an age by gender interaction, showing that men of the same age group as women were less likely to enter a Care Home. Alternatively, as men aged the odds of entry compared to women decreased. There was also a deprivation by gender interaction so that being affluent increased the odds of entry but this was stronger for women than men. Receiving antipsychotics increased the odds but this reduced with age.

Fig. 2 Functional form of entry to a Care Home by age and gender The main factor relating to previous hospitalisation was a strong increase in odds of entry with increasing time spent in hospital in the previous year. A natural log transformation was

found to give the best fit. It was noticeable that previous hospitalisation for cancer was associated with a reduced odds possibly due to dedicated cancer care being positioned outside Care Homes. Although Falls were a significant factor in univariate analysis after adjustment for other factors it was no longer significant and so does not appear explicitly in the model, although many of the medications in the model are associated with falls. Of the community prescribing, a number of types of medication were associated with reduced odds of entry to a Care Home. These were use of ulcer healing drugs, betablockers, lipid regulating drugs and NSAIDS which points to good pharmacotherapy of long term conditions. On the other hand, one of the strongest predictors was receipt of drugs used in dementia (BNF 4.11). Other medications associated with increased odds were antiplatelet agents, hypnotics and anxiolytics, antipsychotics, antidepressants, and Parkinsonsâ&#x20AC;&#x2122;s disease medication. Overall, the total number of prescriptions was a significant predictor of entry to a Care Home independently of the individual medications in table 2. Table 2 Results of logistic regression on Care Home entry over one year of follow-up

Factor

Beta

Se (beta)

Wald

p-value

95% CI

0.588

0.081

52.4

<0.001

1.800

1.535, 2.111

-0.003

0.000

31.1

<0.001

0.997

0.996, 0.998

2.199

0.845

6.8

0.009

9.012

1.721, 47.20

-0.982

0.108

83.2

<0.001

0.375

0.303, 0.463

BNF1.3

-0.327

0.081

16.4

<0.001

0.721

0.615, 0.845

BNF2.4

-0.319

0.090

12.6

<0.001

0.727

0.609, 0.867

BNF2.9

0.220

0.079

7.8

0.005

1.246

1.067, 1.455

-0.314

0.093

11.3

0.001

0.731

0.608, 0.877

BNF4.1

0.247

0.090

7.6

0.006

1.281

1.074, 1.528

BNF4.2

8.723

1.278

46.6

<0.001

6142

501.6, 75209

BNF4.3

0.292

0.090

10.4

0.001

1.340

1.122, 1.600

BNF4.9

0.554

0.183

9.2

0.002

1.739

1.216, 2.488

BNF4.11

1.425

0.189

57.0

<0.001

4.158

2.872, 6.018

-0.311

0.095

10.6

0.001

0.733

0.608, 0.883

0.230

0.039

35.4

<0.001

1.259

1.167, 1.357

0.540

0.024

506.7

<0.001

1.715

1.637, 1.798

Age at baseline Age squared Gender (M vs. F) Deprived vs. Affluent

BNF2.12

BNF10.1.1 Log (Num. of Prescriptions.) Log (Days of most recent hosp.)

Previous Cancer

-0.570

0.268

4.5

0.033

0.566

0.335, 0.956

Age x Sex

-0.030

0.010

8.4

0.004

0.971

0.951, 0.990

Age x BNF4.2

-0.091

0.016

32.1

<0.001

0.913

0.885, 0.942

Sex x Deprivation

0.392

0.177

4.9

0.027

1.480

1.046, 2.095

Constant

-34.4

3.369

104.4

<0.001

Hospitalisation (Yes vs. No)

.000

In order to utilise the full extent of follow-up, a more efficient modelling process is to use survival models. Power was also increased as the number of events was now n = 4717 or 5.6% (table 1). The multiple regression process was repeated with a Cox proportional hazards model and the results are shown in table 3 which gave a similar model but with increased significance and additional factors due to the increased power. Table 3 Results of multiple Cox regression model for time to entry to a Care Home Beta Age at baseline

SE (beta)

Wald

p-value

95.0% CI

0.639

0.035

336.7

<0.001

1.895

1.770

2.028

-0.003

0.000

210.3

<0.001

0.997

0.838

0.374

5.0

0.025

2.311

1.110

4.810

-0.851

0.045

360.9

<0.001

0.427

0.391

0.466

0.877

0.163

29.0

<0.001

2.405

1.748

3.308

0.347

0.124

7.9

0.005

1.415

1.111

1.803

0.267

0.017

256.0

<0.001

1.306

1.264

1.349

BNF1.3

-0.247

0.035

49.8

<0.001

0.781

0.729

0.837

BNF2.4

-0.267

0.038

49.9

<0.001

0.766

0.711

0.825

BNF2.9

0.199

0.034

33.6

<0.001

1.220

1.141

1.305

-0.316

0.040

62.6

<0.001

0.729

0.674

0.788

BNF4.2

7.042

0.684

106.0

<0.001

1143.9

299.4

4370.9

BNF4.3

0.378

0.039

93.4

<0.001

1.460

1.352

1.576

BNF4.9

0.595

0.085

48.8

<0.001

1.813

1.534

2.143

BNF4.11

1.733

0.076

525.0

<0.001

5.655

4.876

6.559

BNF9.1

0.219

0.046

22.4

<0.001

1.245

1.137

1.363

Age squared Gender (M vs.F) Deprived vs. Affluent Log (days in recent hospitalisation) Cerebrovascular Hosp Admission in previous year Log (Num. of prescriptions)

BNF2.12

BNF10.1.1

-0.168

0.039

18.3

<0.001

0.846

0.783

0.913

Age x Gender

-0.014

0.005

9.0

0.003

0.986

0.977

0.995

Age x BNF4.2

-0.075

0.009

76.6

<0.001

0.927

0.912

0.943

0.212

0.079

7.2

0.007

1.236

1.059

1.442

-0.007

0.002

12.3

<0.001

0.993

0.989

0.997

Gender x Deprivation Age x Log (days)

3. Does the derived model have adequate predictive ability and does it accurately predict entry to a Nursing Home using internal validation? The model derived from the logistic regression (table 2) gave an AUROC = 0.89 (95% CI 0.88, 0.90) which is excellent discriminative power. The AUROC is a measure of the modelâ&#x20AC;&#x2122;s ability to differentiate between high and low risk. In comparison the Framingham model for CHD had an AUROC of 0.8. The Hosmer-Lemeshow test is a test of calibration, that is, how accurate is the actual estimate of risk. The Hosmer-Lemeshow test was not significant (p = 0.563) with this model suggesting good calibration or agreement between observed events and events predicted by the model. The results were similar for the Cox model with excellent discriminative power. 4. Can a simple prediction rule be derived for clinical use? The simplest method is based on taking the linear predictor from the model and adjusting the regression coefficients for each factor to a whole number which is then converted into points [6]. The linear predictor from the model XB (the sum of the factors (X) in the model with each multiplied by their regression coefficients or weights (B)). For a logistic model this is straightforwardly achieved by estimating exp (-XB) given the characteristics of a particular person. The probability of entry to a nursing home is then calculated as: p = 1 / (1 + exp (-XB)) This probability can be easily calculated say in an Excel spreadsheet and with additional work programming, embedded in General Practice software such as Vision or Emis. Of course, such an algorithm could also be utilised at the Health Board or CHP level if the data is available for everyone in the population 65+. 5.

Discussion

By utilising Tayside linked routine health care data it was possible to assess associations of the characteristics of those aged 65 or older in relation to their risk of entry to a Care Home in the future. The results were consistent with previous US studies in indicating that being female, having dementia and previous hospitalisations were important predictors [7-9]. Although we did not have direct measures of cognitive decline, use of acetylcholinesterase inhibitors could be seen as a marker for this factor. There were strong associations with medication for dementia and antipsychotic medications as well as age, deprivation and gender. The rate of increase in entry with age accelerated faster for women compared to men and deprived men were less likely to enter a Care Home than women. It is possible that there is competing risks with higher mortality in men depleting the pool of men who had they lived longer, may have gone into a Care Home.

Heavy use of hospitals in the previous year was also associated with a move to a Care Home, although previous Cancer admissions were associated with lower risk, presumably because care would be concentrated elsewhere. Interestingly there were differences in risk factors for Care Home entry compared to unplanned emergency admissions (PEONY) based on the same population dataset [3]. Dementia medication was not a risk factor for unplanned emergency admissions whereas it was a strong risk factor for entry to a Care Home. NSAID prescriptions were a risk factor for emergency admissions in PEONY due to their gastrointestinal toxicity but was associated with lower risk for Care Home entry presumably as they provide pain relief for many long term conditions which might improve mobility and hence capacity for self-care. The fact that there were a number of medications associated with a lower likelihood of entering a nursing home such as statins, ulcer-healing drugs, NSAIDs and beta-blockers suggests that control of certain long term conditions leads to the maintenance of autonomy for longer and there may be scope for more judicious use of these medications. Independently of these individual medications the total number of prescriptions received in the previous year was a strong predictor of entry to a Care Home and this is also consistent with previous studies [10]. Unfortunately, we were unable to accurately determine who was living alone due to difficulties matching addresses although we believe that with further work this could be achieved in future studies. As with all algorithms they tend to perform well on the data from which they were derived and the acid test of any model is how it performs in a new dataset. Hence external validation is the essential next step. Finally, new or existing interventions to maintain autonomy could be evaluated alongside use of the algorithm to improve health in patients and reduce costs to the NHS [11]. References 1. Health Informatics Centre (http://www.dundee.ac.uk/hic/) 2. British National Formulary, BMJ and RPS Publishing, London, 2009. 3. Donnan PT, Dorward DWT, Mutch B, Morris AD. Development and validation of a model for Predicting Emergency admissions Over the Next Year (PEONY): a UK historical cohort study. Archives of Internal Medicine, 2008; 168: 1416-1422. 4. Scottish Care Commission Register held in , Edinburgh (http://www.carecommission.com/index.php?option=com_content&task=view&id=16&Itemid=112 5. Hosmer DW and Lemeshow S. Applied logistic regression. John Wiley& sons, New Jersey, 2000. 6. Sullivan LM, Massaro JM, Dâ&#x20AC;&#x2122;Agostino RB Sr. Presentation of multivariate data for clinical use: The Framingham Study risk score function Statist Med 2004; 23: 1631-1660. 7. Young YC. Factors associated with permanent transition from independent living to nursing home in a continuing care retirement community. J Am Med Dir Assoc 2009; 10: 491-497 8. Smith ER, Stevens AB. Predictors of discharges to a nursing home in a hospitalbased cohort. J Am Med Dir Assoc 2009; 10: 623-9. 9. Gaugler JE, Duval S, Anderson KA, Kane RL. Predicting nursing home admission in the US: a meta-analysis. BMC Geriatrics 2007; 7: 13. 10. Luppa M, Luck T, Weyerer S, Konig H-H, Brahler E, Riedel-Heller SG. Prediction of institutionalization in the elderly: A systematic review. Age and Ageing 2009; 1-8.

11. Beswick AD, Rees K, Dieppe P, Ayis S, Gooberman-Hill R, Horwood J, et al. Complex interventions to improve physical function and maintain independent living in elderly people: a systematic review and meta-analysis. The Lancet 2008;371(9614):725-735.

Importance to NHS and possible implementation

Maintaining physical and cognitive function is important to the elderly and could save costs to the NHS. Implementation of such a predictive model requires the steps outlined below for future research. It would be important to provide a simple and user-friendly interface for the model if it was to be used in practice. 7.

Future research Further work would be required to develop the models with the aim of adding social function factors that may be important in the change from residence in their own homes to a Care Home. Competing risks models should be explored as the differential mortality in men compared to women would change the proportions available for entry to a Care Home. Programming expertise is required to convert the statistical model into a user-friendly algorithm embedded in general practice systems. This could follow the exemplar of PEONY which has now been implemented in NHS Tayside. External validation would be essential for transportability of the algorithm. Finally, evaluation in the form of interrupted time series or randomised trials would determine whether use of the algorithm is feasible in practice, usable, and makes a difference to patient outcomes and NHS outcomes such as costs.

Dissemination The results will be presented at international conferences such as the SHIP biennial conference in St Andrews in September, 2011. A paper will be submitted to a high impact journal such as the BMJ.

Research workers

Dr Karen Barnett derived the datasets and carried out the statistical analyses. 10.

Financial statement

This has been sent separately by the University of Dundee finance department. 11.

Executive summary (Focus on Research)

See separate attachment.

Appendix 1

Potential Drug prescription predictors based on British National Formulary chapter (2005)

Gastro-intestinal – ulcer healing drugs (BNF chap 1.3); CVD – diuretics (BNF chap 2.2 - CHF) - beta-blockers (BNF chap 2.4) -anti-hypertensives (BNF chap 2.5) -nitrates (BNF chap 2.6) -anticoagulants (BNF chap 2.8) -antiplatelets (BNF chap 2.9) -antifibrinolytic drugs (BNF chap 2.11) -lipid regulating drugs (BNF chap 2.12) Asthma/COPD – LABA, SABA and / or inhaled corticosteroids (BNF chap 3.1, 3.2, 3.3) – treat as one category BNF chap 3 CNS

-hypnotics and anxiolytics (BNF chap 4.1) -antipsychotics (BNF chap 4.2) -antidepressants (BNF chap 4.3) -analgesics (BNF chap 4.7) -antiepileptics (BNF chap 4.8) -anti-Parkinsonian (BNF chap 4.9) -drugs used in substance dependence (BNF chap 4.10) -anticholinesterases (BNF chap 4.11 - dementia)

Antibacterial drugs (BNF chap 5.1) Diabetes – oral sulphonylureas, metformin, insulin (BNF Chap 6.1) Osteoporosis (BNF chap 6.6) Anaemia (BNF chap 9.1) NSAIDS (BNF chap 10.1.1);

Appendix 2

Univariate Associations with Predictor Variables – Logistic Regression

Variable P value Demographic Variables Age at Baseline 0.00 (Continuous) Sex (M=1, F=0) 0.00 HB SIMD10 0.00 (Continuous) Prescription Variables Log of Total No. 0.00 Scripts (continuous) BNF1.3 0.028 BNF2.2 0.000 BNF2.4 0.063 BNF2.5 0.745 BNF2.6 0.411 BNF2.8 0.006 BNF2.9 0.000 BNF2.11 0.196 BNF2.12 0.000 BNFCh3 0.718 BNF4.1 0.000 BNF4.2 0.000 BNF4.3 0.000 BNF4.7 0.000 BNF4.8 0.001 BNF4.9 0.000 BNF4.10 0.286 BNF4.11 0.000 BNF5.1 0.000 BNF6.1 0.821 BNF6.6 0.000 BNF9.1 0.000 BNF10.1.1 0.003 Hospital Variables Total No. 0.000 Hospitalisations Total Days In 0.000 Hosp Length Of stay – 0.000 Most Recent Hospitalisation Time from 0.000 Baseline to Previous hosp

Odds Ratio

Lower 95%CI

Upper 95% CI

1.15

1.14

1.16

0.61 1.11

0.53 1.08

0.70 1.13

One Year History 1.628

1.529

1.733

1.167 1.925 0.861 0.977 1.060 1.437 1.994 3.752 0.702 1.034 2.744 8.721 2.117 1.921 1.643 4.890 0.733 13.719 1.578 1.027 1.900 3.233 0.768

1.017 1.695 0.736 0.847 0.923 1.109 1.756 0.506 0.602 0.862 2.380 7.088 1.823 1.692 1.222 3.609 0.414 10.147 1.389 0.813 1.499 2.722 0.647

1.340 2.185 1.008 1.126 1.216 1.863 2.265 27.808 0.819 1.241 3.165 10.730 2.459 2.181 2.209 6.625 1.298 18.548 1.793 1.298 2.407 3.840 0.912

1.201

1.166

1.237

1.031

1.029

1.034

1.029

1.027

1.032

0.995

0.994

0.995

admission. SMR01 Variables â&#x20AC;&#x201C; Cause of Most Recent Admission CHD 0.004 1.772 1.204 Cerebrovascular 0.000 7.827 5.422 CHF 0.013 2.823 1.249 Cancer 0.209 1.364 0.841 Diabetes 0.034 4.668 1.124 Respiratory 0.000 3.212 2.161 Falls 0.000 7.096 5.389

2.608 11.298 6.381 2.211 19.396 4.776 9.343

GUIDANCE ON FINAL REPORTS (FORM 4)

Appendix A

CONTENT / FORMAT A final report of approximately 2,500-3,500 words is required at the end of the project. Twenty-five copies of the final report should be submitted, along with an electronic copy (emailed to: csograntapplications@scotland.gsi.gov.uk). The format should be as shown in Appendix A (attached) and should include both a scientific summary and a ‘Focus on Research’ lay executive summary (see Appendix B attached). The ‘Focus on Research’ summary will be posted on the CSO website as soon as possible after the report is passed by the relevant Committee.

PURPOSE The purpose of providing a Final Report is to account for the monies spent, and to enable the relevant Committee to consider whether the project fulfilled its original aims, analysed the data to a satisfactory standard and presented the findings in a scientifically rigorous and justifiable way. The Committee will also wish to see evidence that any policy and / or services implications have been set out, and that dissemination is properly planned with evidence of progress.

CHANGES FROM ORIGINAL APPLICATION Given the focus on whether the project fulfilled its original aims, it is vital that any departures from the original aims or methods are fully justified and that a clear and transparent account is given of any difficulties faced (staffing, recruitment etc) and the steps taken to minimise the impact of any unforeseen difficulties. Such departures should have been agreed in advance with the CSO Research Manager and this will be evident from the summary of each project’s history that is presented to the relevant committee. This includes any difficulties reported to CSO by the investigators which may have affected the timing of completion or the extent to which the project was able to achieve its objectives. In addition, BTRC applicants will be sent a self assessment form (see Appendix C) to complete which details departures from the original programme of work, in terms of methodology, recruitment and staffing.

LENGTH The word count for the final report should be between 2,500 and 3,500 excluding tables. This is the recommended guideline for both qualitative and quantitative reports and a word count should be included to demonstrate this guideline has been adhered to. Only in exceptional circumstances will longer reports be accepted. The final report is not a substitute for other forms of dissemination, but is specifically for the purpose of accounting for the use of funds and this word length is sufficient for the purposes of judging the report. This length is adequate for both qualitative and quantitative reports to present sufficient data to illustrate the main findings.

ADDITIONAL MATERIAL Appendix material and published papers may also be included with the final reports. Six copies of these will be required. Published papers will not be acceptable in lieu of a final report. Publications arising from the project should acknowledge the CSO and should be sent to CSO as soon as they are available. The appendix material will not always be studied in detail – especially if there is a great deal of it – so investigators should ensure that the key points are communicated in the body of the report.

ASSESSMENT PRIOR TO COMMITTEE MEETING Before final reports are included in the agenda, they are reviewed by the Research Manager. If it is clear that a report is not in the right format, or fails to provide an adequate account of how the objectives of the study have been addressed, the Research Manager may return it to the Chief Investigator (CI) for revision. CIs that have concerns that their report may not meet the requirements for final reports are advised to send a draft to the relevant Research Manager as soon as they have agreed the content with their co-investigators. The Research Manager will let the CI know whether the report conforms to the guidelines and, if necessary, correspond with the CI until the report is in a suitable form to go to the Committee. This should result in fewer reports being rejected or returned by the Committee.

ASSESSMENT AT COMMITTEE All Committee members receive a copy of every final report. The Chair, speaker(s) and scorer(s) also receive a copy of the application and any other relevant papers.

Health Services Research Committee For the Health Services Research Committee, the Chair and three members are asked to score each final report prior to discussion by the Committee. The criteria for assessment and the scoring system are set out in Appendix D. As can be seen from the Appendix, this Committee does not distinguish ‘Good’ reports from ‘Satisfactory’ reports. The grade ‘Satisfactory’ therefore covers the vast majority of reports and indicates that the research delivered on its objectives. ‘Excellent’ reports are rare, and indicate an outstanding project which has gone beyond what could reasonably have been anticipated and which has also been delivered on time and within budget. Reports which are not accepted are returned for rewriting and the reasons for returning the report are explained. Revised reports are considered at the next meeting of the Committee. At this stage they may be graded ‘Satisfactory’, they may be accepted without being graded satisfactory, or they may be graded ‘Unsatisfactory’. The Health Services Research Committee does not perform the role that would be expected of a specialist reviewer of an article submitted to a journal.

Biomedical and Therapeutics Committee For the Biomedical and Therapeutic Research Committee, three members are asked to score each final report prior to discussion by the Committee. The criteria for assessment and the scoring system are set out in Appendix E. ‘Excellent’ reports are rare, and indicate an outstanding project which has gone beyond what could reasonably have been anticipated and which has also been delivered on time and within budget. Some reports are accepted and not scored, for example when the project is terminated without an outcome.

ACTIONS FOLLOWING ASSESSMENT Reports which are accepted by the Committee are disseminated by putting the ‘Focus on Research’ summary on to the CSO website. CSO also ensures that details of completed reports are made available to relevant parties within the Scottish Executive. CSO has introduced a wider (paper based) dissemination of the ‘Focus on Research’ summaries to key stakeholders. If a resubmitted report is marked ‘unsatisfactory’ the final payment on the project will be withheld and the Chief Investigator will not be able to apply for further CSO funds for a period of 12 months from the date of the Committee discussion. Any specific comments highlighted by the Committee will be communicated to the investigators in writing after the meeting at which the report was considered.

Appendix B Guidance on Focus on Research Summary The aim of Focus on Research summaries of final reports is to improve access to your research and to stimulate interest in the final report. These summaries can also be used by CSO to disseminate information concerning the projects we fund throughout the Health Service in Scotland. Preparation of the summary (and indeed the whole report) should also encourage consideration of how the findings may contribute to the development of health service policy and practice and to formulate thoughts about the direction the research should take in the future. The text of the summary should be easily understandable and describe the main results and potential practical implications. The executive summary is intended to be read without reference to the report as a whole and therefore some constraints apply: • • • • • • •

the summary should not contain references; any abbreviations or specific terms should be explained; a contact for obtaining copies of the full report must be given; the text should be readily understandable (written to be understood by the layperson); it should focus thoughts on how the study findings fit into the bigger picture; it must be presented in the required format; it must be returned to CSO in electronic format (along with the final report).

Format A standard format is required as on the template which can be downloaded from the website along with sample summaries at http://www.sehd.scot.nhs.uk/cso/ApplyingForFunding/ResponseMode/ExecSummNew.htm Please note: • • • • • •

the section headings are given and all should be included, although the relative sizes of the sections is flexible (the template is in 2 columns, and the text will automatically wrap to the next column); click on the line below a section heading to insert the text; please leave one free line between sections; please use the fonts as set (Verdana, 12 point for headings and 10 point for text) and do not use smaller font sizes in an attempt to include more information; the summary as a whole must not exceed the one page given in the template; if your word processing package is more recent than Word 2000, please save in Rich Text Format (rtf).

Sections Aim: a short description of what the study set out to achieve including background if required for the following text to be understood. Project outline/methodology: how the study was carried out, including criteria for participants, tests carried out etc. Key Results: highlight the significant findings of the study. Conclusions: short statement of the outcome of the study - did it achieve its aims?

What does this study add to the field: this should summarise what was previously known and how these data add to the body of knowledge. Implications for practice or policy: this should describe whether this study could potentially lead to changes in practice or policy. Where to next: this should state what further research naturally follows on from the completed study. If the Focus on Research page of the report is not in the required format, is not easily understandable, does not stand alone from the report or is not received in electronic format, it will be returned for editing. A member of CSOâ&#x20AC;&#x2122;s Public Involvement group will also comment. Failure to revise the summary may result in delays in the final payment of the grant being issued. Executive summaries will be posted on our website after consideration at the relevant advisory committee meeting. CSO conducted a review of journal policies towards publication of summaries on the web and this suggested that most leading journals would not reject papers for this reason (this is not unlike having an abstract published at a Conference). Any concerns that posting of the Summary on the website may lead to rejection of a paper by a journal should be directed to the relevant Research Manager.

Appendix C

BTRC Self Assessment Form Project Reference Number: Chief Investigator:

1. Has the project fulfilled its original aims? Yes In part No If No or In part, answer 1a, b 1a. Are any departures from those aims justified? 1b. Were any departures agreed with CSO?

Yes Yes

In part In part

METHODS

No No

2. Did the project use the methods originally set out? Yes In part If No or In part, answer 2a, b

2a Are departures from those methods justified in the report?

Yes

2b Were any departures agreed with CSO? Yes

In part

3. Did the project encounter technical difficulties? Yes (i.e. in the application of methods) If 'Yes', answer 3a 3a Were all reasonable steps taken to overcome them? Yes Sample size (if appropriate) Category of subject Original sample size • • • 4. Was an adequate sample size achieved? If No, answer 4a, b, c

In part No

No In part

Sample achieved

Yes

4a Were all reasonable steps taken to achieve an adequate sample size? Yes

4b Was CSO informed of problems with recruitment?

Yes

4c Was the sample size achieved sufficient to answer the question? Yes

5. Did the project encounter staffing difficulties? If 'Yes', answer 5a

Yes

5a Were all reasonable steps taken to minimise their impact? 6. Original end date = Revised end date =

No Yes

Appendix D

HSRC Final Report Assessment Form Agenda No: Project title:

Member's Name:

Please read the notes on scoring before completing the form The following questions should be answered in relation to the conduct of the research 1. Has the project fulfilled its original aims? If No or In part, answer 1a 1a. Were any departures from those aims justified?

Yes

In part

Yes

In part

2. Did the project use the methods originally set out? If No or In part, answer 2a 2a Were any departures from those methods justified?

Yes

In part

Yes

In part

3. Was an adequate sample size achieved? Yes No If No, answer 3a 3a Were all reasonable steps taken to achieve an adequate sample size? 4. Did the project encounter technical difficulties? If 'Yes', answer 4a

Not applicable Yes

Yes

4a Were all reasonable steps taken to overcome them?

Yes

In part

5. Were the data analysed to a satisfactory standard? Yes

In part

The following questions should be answered in relation to the report 6. Are sufficient data presented to report the findings?

Yes

In part

7. Are the conclusions supported by the data?

Yes

In part

8. Are policy/service implications set out?

Yes

In part

9. Is the report clearly written, avoiding unnecessary jargon?

Yes

10. Is there a clear dissemination plan with evidence of progress?

Yes

Not applicable

In part In part

No No

The following questions should be answered in relation to project management 11. Did the project encounter staffing difficulties? If 'Yes', answer 11a

Yes

11a Were all reasonable steps taken to minimise their impact? Yes 12. Was the report produced on time?

Yes

3-6 months late

6-12 months late

over 12 months late

SCORE

Excellent Satisfactory Returned Unsatisfactory

COMMENTS

SCORING 1. Returned

Any report which does not achieve a 'Yes' or 'In part' at all of the following questions, 1 (or 1a), 2 (or 2a), 5, 6 and 7, will be returned to the investigator for rewriting.

2. Unsatisfactory

Any report which, even after rewriting, fails to achieve a 'Yes' or 'In part' at questions 1 (or 1a), 2 (or 2a), 5, 6 and 7, will be graded unsatisfactory. The penalty is that no application for funding may be made by the CI to the CSO until a satisfactory report is received.

3. Satisfactory

Reports which meet the minimum requirements, but do not stand out as excellent.

4. Excellent

Reports which achieve a yes on items 1-11 (including 1a, 2a, 3a, 4a, and 11a where relevant) and demonstrate high quality work on all aspects of conduct of the research and reporting. This grade is intended to identify the small minority (c10%) of reports which are clearly outstanding.

NOTES

(1) Reports over 12 months late will be graded Satisfactory (late) or Excellent (late) (2) The shaded items (11-12) will be completed by the office.

Appendix E

BTRC Final Report Assessment Form Agenda number Project Title

Please circle appropriate response Does the report answer the projectâ&#x20AC;&#x2122;s questions?

Yes

Has the proposed methodology been followed?

Yes

Does the study represent value for money?

No No

Yes

Does the report contain appropriate proposals for dissemination?

Yes

No No

Please circle appropriate score Excellent

Good

Satisfactory

Unsatisfactory

Further comments (these may be passed on to the grantholders)

Memberâ&#x20AC;&#x2122;s name