Author's response to reviews Title:Emergency ambulance service involvement with residential care homes in the support of older people with dementia: an observational study Authors: Sarah Amador (s.amador@herts.ac.uk) Claire Goodman (c.goodman@herts.ac.uk) Derek King (d.king@lse.ac.uk) Ina Machen (i.machen@herts.ac.uk) Natasha Elmore (nb382@medschl.cam.ac.uk) Elspeth Mathie (e.j.mathie@herts.ac.uk) Steve Iliffe (s.iliffe@ucl.ac.uk) Version:3Date:14 May 2014 Author's response to reviews: see over
University of Hertfordshire Hatfield AL10 9AB UK tel +44 (0)1707 284000 fax +44 (0)1707 284115 herts.ac.uk Editor BMC Geriatrics 14/05/14 Dear Editor, EMERGENCY AMBULANCE SERVICE INVOLVEMENT WITH RESIDENTIAL CARE HOMES IN THE SUPPORT OF OLDER PEOPLE WITH DEMENTIA: AN OBSERVATIONAL STUDY Thank you for your helpful comments on the above manuscript received April 24 2014. We are pleased to resubmit the revised manuscript for your review. Please find below a point-by-point response to your comments which we hope will satisfactorily address all of your concerns. We have taken care to address the recommendations and be explicit about the limitations of the data. We believe this is an important paper because it is the first UK study that has addressed use of emergency ambulance services by people with dementia in care homes. Previous work has relied on hospital episode data which does not capture residents who are not conveyed or admitted to hospital. We are grateful to the reviewers and the editor for their careful and constructive consideration of this paper and believe it has been improved through this process. We look forward to hearing from you in due course. Yours sincerely, Professor Claire Goodman, Dr. Sarah Amador, Dr. Derek King, Ina Machen, Natasha Elmore, Dr. Elspeth Mathie, and Professor Steve Iliffe
REVIEWER 1 COMMENTS I enjoyed reading this redrafted paper and I think it contains some useful data. Apart from the conclusion, which I feel does not coherently relate to the rest of the paper, I have only very minor suggestions. Thank you for your very helpful comments on this and the previous version of the paper Discretionary: a. Remove the words "by older people with dementia in residential care homes" from sentence 2 para x of the introduction These words have been removed. b. Change "using" to "from" as the fifth word in the first sentence of the methods This has been corrected. c. Remove the words "predictors of contact with emergency ambulance services included" and replace with "we modeled" This has been amended. d. The words "and we proceed with caution as regards interpretation of results" can be removed from page 10. This has been removed. Minor: Perhaps you could provide details of how the logistic regression was done in the methods section? Were all potential variables entered into the model or were they tested in a stepwise fashion? Thank you. The analyses were adjusted by patient-related variables chosen in advance of the analyses and entered together. The second to last paragraph has been amended as follows. The logistic regression analyses adjusted for factors potentially associated with emergency ambulance service use. Based on empirical findings on emergency department use by long-term care residents as reviewed above, the adjustment variables selected were the age of the resident, gender, length of residency in the care home, number of co-morbidities, admission route into the care home, use of other services (i.e. general practitioner, district nurse) and the care home itself. All covariates were entered into the model. We predicted that emergency ambulance use is related to case complexity as measured by number of co-morbidities and contacts with general practitioners and district nurses, as well as age. The goodness of fit of the logistic regression model was assessed by firstly, comparing the full model with a constant only model to determine the level of significance of the set of independent variables; and further by assessing the Nagelkerke s R-squared statistic [17] and the percentage of observations in which the model correctly predicted the dependent variable. Major I believe the conclusion again needs to be rewritten, to tie in with the rest of the paper. There is no data presented regarding 12 month mortality, end of life care etc. of the residents in this study, so it seems wrong to make end of life care the focus of the conclusion. The conclusion to be drawn from this study is ambulance usage in homes of the type studied is high, largely for ACS conditions and to some extent predictable.
We agree. The conclusion has been rewritten as follows: Emergency ambulance service use by older people with dementia in residential care homes is high, associated with ambulatory care sensitive conditions and to some extent predictable. Emergency service involvement with older people with dementia merits further examination. Future interventions to reduce inappropriate use of services should consider how the wider context of care as well as a resident health characteristics influence decision making to call out emergency services and admit older people to hospital. REVIEWER 2 COMMENTS I am happy that the authors have addressed my previously stated concerns with their amendments to the article. Thank you for your very helpful comments on the previous version of the paper. EDITOR'S COMMENTS Additional Editorial Comments: "I m satisfied with all the responses to my comments except for the following one: Methods Page 5, paragraph 2, first sentence - What specification tests were used to assess goodness of fit for the logistic regression model? Results regarding specification tests to assess goodness of fit for the logistic regression model have been included in the beginning of the last paragraph of the results section as follows: A test of the full model against a constant only model was statistically significant, indicating that predictors as a set, effectively distinguished between residents who had contact with emergency services and those who did not (chi square = 40.037, p < 0.001 with df = 15). Nagelkerke s R2 of.379 indicated a relatively good relationship between prediction and grouping. Prediction success overall was 80% (75.4% for no contact and 84.1 for contact with emergency services). While this is helpful, information from additional specification tests is needed to determine whether the model?s fit is adequate. Please provide the test statistic and p value for the following tests: Pregibon?s linktest, Ramsey?s RESET test, the Hosmer-Lemeshow test, and the C statistic (area under the ROC [receiver operating characteristics] curve) at the end of Table 4. These are all standard specification tests. The first three are especially important in ascertaining the functional form of the model and whether polynomials or interaction terms are required. At this point I have no idea whether any of the continuous variables should be modeled as polynomials. With regard to addressing the specification tests, we suggest that the new statistical detail could act as a distraction from the key messages of the report. Interaction terms were not selected due to the relatively small number of observations. It is established that in order to detect interactions many more observations are required compared to the numbers needed for straightforward main effects modelling [please see Brookes ST, Whitely E, Egger M, Davey Smith G, Mulheran PA, Peters TJ. Subgroup analyses in randomized trials: risks of subgroup-specific analyses; power and sample size for the interaction test. Journal of Clinical Epidemiology 2004, 57:229-236)]. For instance, in a study of over 2,100 stroke patients numerous main effect relationships were significant but no interaction terms [please see Smeeton NC, Corbin
DOC, Hennis AJM, Hambleton IR, Rose AMC, Fraser HS, Heuschmann PU, Wolfe CDA. A comparison of outcome for stroke patients in Barbados and South London. International Journal of Stroke 2011, 6(2): 112-117]. Similar sample size issues apply in the consideration of polynomial terms. New Comments 1. I m confused by what the dependent variable of the model actually is. In Table 4 it says contacts. The last three sentences of the Background section mentions emergency ambulance service use. The third paragraph of the Methods talks about call-outs. The fourth paragraph uses the terminology use and call-outs. Please define emergency ambulance service use, call-outs, and contacts as early as possible in the paper, and be as consistent as possible throughout the paper in their use. There should be a sentence in the Methods that uses the words dependent variable so the reader will know exactly what is in the model. At this point I m not sure whether the logistic regression includes people who were not conveyed to the hospital or not. Thank you, we agree that the dependant variable needs to be made explicit and consistency maintained throughout the paper. The term call-outs has been replaced with contacts throughout the paper. The term contact is defined at the beginning of the third paragraph of the methods section as follows: Emergency ambulance service use was defined as the number of contacts with an emergency ambulance. The terms dependent variable has been included as follows in the second to last paragraph of the methods section: The dependent variable is the number of contacts with an emergency ambulance. 2. For the logistic regression, joint Chi square tests should be reported for the 4 admission routes together. Otherwise you don t know if admission route is significant as a group. Similarly, this should be done for the 5 care homes. Again, unless you do this you don t know if care homes are significant as a group. Thank you for your comment. We have not however, taken this approach in order to avoid multiple testing, which increases the risk of false positive findings. 3. What kind of standard errors was estimated for logistic model? Did it take account of patients being clustered within care homes? This should be mentioned in the Methods or as a note at the bottom of the table. Thank you for your comment. The use of multilevel modelling seems to be suggested here. We have not considered multilevel modelling as the coefficient estimates for logistic regression can be seriously biased if the number of clusters (i.e. homes) is less than 10 [see Austin PC. Estimating multilevel logistic regression models when the number of clusters is low: A comparison of different statistical software procedures. International Journal of Biostatistics 2010, 6(1): article 16]. 4. I m a little confused about Table 2. The title says baseline characteristics and service use. What period do the hospital services, emergency ambulance services, community health care services, and primary health care services cover? Is it prior to the baseline characteristics of age, gender, etc.? Or does it start with entry into the study?
Thank you for your comment. We agree that Table 2 is unnecessary as results have already been reported in full in the text (please see c Paragraph entitled Baseline characteristics and service use in the results section). As a result, Table 2 has been entirely removed. 5. Table 4 includes Number of General Practitioner contacts and Number of District Nurse contacts. I don t see those included in Table 2. Please see response to comment 4 above. 6. I m somewhat concerned about the relatively small number of observations per explanatory variable in your model. A general rule of thumb for logistic regression is 15 observations per variable. Your model has 133 observations and 15 variables, so there are 8.9 observations per variable. As admission route is something that occurred in the past, you might exclude those variables from your model. Doing that would increase the number of observations per variable to 12.1. Another approach would be to do a separate model for each explanatory variable (age, gender, etc.), including (controlling for) the care homes in each model since they re so important. I ve done similar analyses when I ve had small sample sizes. Thank you for your comment. Doubts have been expressed about the rule of 15, and some statisticians believe that using 10 or even fewer observations per explanatory variable is acceptable [please see Vittinghoff E, McCulloch CE. Relaxing the rule of ten events per variable in logistic and Cox regression. American Journal of Epidemiology 2007, 165(6): 710-718]. 7. The statistical software used should be mentioned at the end of the Methods. Thank you for your comment. The statistical software used has been mentioned at the end of the methods section as follows: Data were analysed using Stata 10.1 [17].