Unplanned nursing home admission among discharged polymedicated older inpatients: a single-centre, registry-based study in Switzerland

Objective To investigate patient characteristics and the available health and drug data associated with unplanned nursing home admission following an acute hospital admission or readmission. Design A population-based hospital registry study. Setting A public hospital in southern Switzerland (Valais Hospital). Participants We explored a population-based longitudinal dataset of 14 705 hospital admissions from 2015 to 2018. Outcome measures Sociodemographic, health and drug data, and their interactions predicting the risk of unplanned nursing home admission. Results The mean prevalence of unplanned nursing home admission after hospital discharge was 6.1% (n=903/N=14 705). Our predictive analysis revealed that the oldest adults (OR=1.07 for each additional year of age; 95% CI 1.05 to 1.08) presenting with impaired functional mobility (OR=3.22; 95% CI 2.67 to 3.87), dependency in the activities of daily living (OR=4.62; 95% CI 3.76 to 5.67), cognitive impairment (OR=3.75; 95% CI 3.06 to 4.59) and traumatic injuries (OR=1.58; 95% CI 1.25 to 2.01) had a higher probability of unplanned nursing home admission. The number of International Classification of Diseases, 10th version diagnoses had no significant impact on nursing home admissions, contrarily to the number of prescribed drugs (OR=1.17; 95% CI 1.15 to 1.19). Antiemetics/antinauseants (OR=2.53; 95% CI 1.21 to 5.30), digestives (OR=1.78; 95% CI 1.09 to 2.90), psycholeptics (OR=1.76; 95% CI 1.60 to 1.93), antiepileptics (OR=1.49; 95% CI 1.25 to 1.79) and anti-Parkinson’s drugs (OR=1.40; 95% CI 1.12 to 1.75) were strongly linked to unplanned nursing home admission. Conclusions Numerous risk factors for unplanned nursing home admission were identified. To prevent the adverse health outcomes that precipitate acute hospitalisations and unplanned nursing home admissions, ambulatory care providers should consider these risk factors in their care planning for older adults before they reach a state requiring hospitalisation.


REVIEWER
Marashi-Pour, Sadaf NSW Health, Bureau of Health Information REVIEW RETURNED 29-Oct-2021

GENERAL COMMENTS
Thank you for providing your paper for review. This work has looked at factors associated with unplanned institutionalization following acute hospitalisations among older patients and could be of great value aiming to improve care among older patients. The study has some limitations that need to be clearly addressed and could benefit from few further analyses. Please see below my detailed comments: • Not having the information regarding hospitalizations, transfers and rehospitalizations to other hospitals, except for the Valais hospital could be a big limitation for this study. It is possible that a prior hospitalisation in another hospital (e.g. a prior surgery) have had an effect on the unplanned institutionalization or a patient might be discharged to nursing home following a transfer to another hospital for a short service. It is not clear how these issues have been taken into account. For example, to reduce its effect, patients coded as transferred in/transferred out (if this information was available at admission/discharge) could be excluded from the analyses, after checking any potential bias that their exclusions might cause If linked data including the entire patients' pathway is not used it needs to be highlighted as a limitation of the study.
• The unplanned institutionalizations need to be defined more clearly based on the available data elements. For example, was a code available in the data regarding the discharge position of the patients following their acute admissions at Valais hospital which was used?
• At page 7 row 200, It is not clear why the modelling analysis included three of the health status variables (e.g. based on univariate analyses?).
• It is not indicated how the multivariable models were developed (e.g. any stepwise modelling approaches used?) • It is mentioned that GEE regression model was used but some details are not provided, including why the data was correlated (e.g. repeated hospitalisations for the same patients?), or what is considered in the model as the within-group correlation structure.
• It is not clear if there were any deaths and if yes how deaths were taken into account. If this information wasn't available and /or wasn't taken into account, it needs to be highlighted as one of the limitations of the study. Some of the observed associations might be explained by deaths, if not taken into account using methods such as competing risks regression models.
• In the results section at page 9 row 264, it is not clear what "g" is referring to.
• The results mostly only include percentages, and lacking numbers. I think it's better that numbers be added to the results. • In the results section end of page 9, the odds ratios aren't described properly.

GENERAL COMMENTS
The Introduction should be shorter and more concise, some paragraphs should be in the Discussion section and not in the Introduction.
The Methods should also be made more clear. The description of the population sample should be in the beginning of the Results section and not in the Methods (make it harder to read).
In the Results you repeat almost all the information that are in the tables and figures in the text (they should be complementary and not repetition of the same information. In Our study fell within the framework of a nationally funded research programme in Switzerland (NRP 74). Specific clinical syndromes or medical diagnostics did not fall within the research programme's focus, nor that of this publication, but they could be the subjects of further publications.

2)
We used retrospective data from a longitudinal, population-based hospital registry of electronic health records. Despite this limitation, the use of routinely collected data is considered an added value to predictive analysis and should not be considered as a gap In response to this comment, we have added the following sentence in the Data analysis strategy section: "A multiple bivariate logistic regression analysis was conducted using crosstabulations to investigate whether the sociodemographic, health and drugs data (more than one independent variable) significantly predicted unplanned nursing home admission from 2015 to 2018 (our single dichotomous outcome).

Data analysis strategy
Lines 236-239 6. It is mentioned that GEE regression model was used but some details are not provided, including why the data was correlated (e.g. repeated hospitalisations for the same patients?), or what is considered in the model as the within-group correlation structure.
To answer this request, we have completed the manuscript as follows: "In a second stage, a series of generalised estimating equations (GEE or population-averaged logistic regression models) were computed to predict how sets of predictors influenced the probability of unplanned nursing home admission. The variables included were derived from the significant associations between sociodemographic characteristics, clinical and medical conditions and unplanned nursing home admission ( We have updated our results by adding absolute numbers to all the percentage prevalence values reported in the manuscript.
Whole manuscript 10. In the results section end of page 9, the odds ratios aren't described properly.
We agree with the reviewer that the data (averages and not odds ratios) were not presented adequately. To avoid ambiguity, we have completed this section as follows: "On average, older adults whose discharge to a nursing home was unplanned had more prescribed drugs

Results
Lines 301-304 than those returning home [10.9 (SD = 3.9) drugs vs 8.9 (SD = 3.2)]." 11. Table 3 at page 21 needs reformatting as currently is hard to read Thank you for your comment. Table 3 presents a summary of the predictive analysis, and this is why the presentation is less standardised. To facilitate readability, we have distinguished between the risk factors and protective factors of unplanned nursing home admission. Furthermore, the different factors are classified in decreasing magnitude of odds ratio. Table 3 (now a  supplementary table) 12. Page 10 at results section lines 299-301 is not very clearly written and perhaps some rewording could be helpful.
We believe that the answer to question 6 (in the data analysis section) brings more clarity to this sentence in the results section. Nevertheless, we have tried to reformulate it as follows: " Figure 3 and Supplementary We have revised and shortened the whole introduction as suggested by the reviewer.

Introduction
The Methods should also be made more clear. The description of the population sample should be in the beginning of the Results We have made the amendment as suggested. A new section-Population description-now appears at the beginning of the Results.

Dataset customizing for predictive analysis
Page 7-8 section and not in the Methods (make it harder to read).

Results
Page 9-10 In the Results you repeat almost all the information that are in the tables and figures in the text (they should be complementary and not repetition of the same information.
We have simplified the text to avoid unnecessary repetition with the figures and tables. Nevertheless, since the journal is also intended for clinicians, we have intentionally left some text in addition to Figures 2 and 3, to support the interpretation and implementation of the GEE logistic regression model's findings.

Results
In table 1 you should mention the number of people in each group (eg. Female and Male...) and not only the percentage so we can better understand the size difference between them. You also should mention the values obtained even the ones that were not statistically significant (the same for table 2).
We agree with your suggestion, and we have completed Tables 1 and 2 with the absolute numbers beside the percentages. Table 1  Table 2 As mentioned before, there shouldn't be repeat information. Therefore, table 3 or figures 1 and 2 should be removed since it is the same information in both places. You can choose to present the information between tables or figures, but do not present the same information in both.
To avoid this redundancy, Table 3which presents a summary of the predictive analysis-is now a Supplementary File.

GENERAL COMMENTS
Thank you for providing your revised paper for review. Please see below my further comments: • I think currently the paper is too long and making it a little shorter could make it easier to follow. For example, maybe more emphasis on the adjusted associations in the text rather than the univariate results could help.
• I think the definition of the unplanned nursing home admission and patients included in the analyses need more clarification. For example: How planned admissions to nursing home among the included patients could be distinguished? It is mentioned in the responses to my previous comments that patients returned home/deaths were excluded from the entire analyses? and all patients followed a home to hospital to long-term residential care facility. I think it is worth making this clearer in the paper including what a long-term residential care facility includes (e.g., one is nursing home which is the outcome of interest). And what was the reason for any of the exclusions, the number excluded, and any potential biases caused by the exclusions needs to be mentioned and discussed. • Following the above comment, clarifications needed regarding the following sentence at line 252 of the results section: "On average, older adults whose discharge to a nursing home was unplanned had more prescribed drugs than those "returning home" [10.9 (SD = 3.9) drugs vs 8.9 (SD = 3.2)]". • Not having access to the linked data and data from other hospitals and any potential biases caused by these and by any of the exclusions need to be highlighted more clearly in the discussion section.
• I think the data analysis strategy section need some rewording to become clearer and shorter. For example: -Please replace the word bivariate in the paper with the word univariate to more clearly refer to the unadjusted analysis conducted. For example, at line 199 something like "univariate analysis using logistic regression models was conducted to investigate…" instead of "multiple bivariate logistic regression analysis was conducted using cross-tabulations to investigate…". Including the odds ratios in these tables could also be useful. And it seems that variables significantly associated with unplanned nursing home admission in the univariate analysis, were used to develop the multivariable models using GEE models.
-Line 207: this sentence is not very clear: "This baseline model was completed using the drugs prescribed to older inpatients who underwent unplanned nursing home admission." Does it mean that the base line model was completed by adding to it drugs that were found significantly associated with unplanned nursing home admission based on the univariate analysis? Did adding drugs to the model make any changes in the associations observed and reported in the baseline model in the previous figure/supplementary table? -How authors have ensured that their final multivariable model is not overfitted? How many parameters were included in the final multivariable model? -As is also indicated in the paper the GEE method is used to estimate population-averaged estimates. Please delete lines 213-216 (or please modify).
• Line 242: with five or more diseases • Line 246, for investigating the association between number of drugs prescribed and unplanned nursing home admission in the univariate analysis authors could use logistic regression consistent with other univariate analyses performed. • In the discussion section line 324, please modify the wording to make it clear that the comparison being mentioned is related to the percentages in the unadjusted analysis, as with the current wording (i.e., tenfold higher risk) it could be confused with relative risks/odds ratios and the adjusted analyses.
• Study strengths and limitations, line 378: authors have indicated that their findings regarding the single hospital, included in their study, could be generalised to other regions of Switzerland. Please include in the paper how authors came to this conclusion. Any representativeness analysis been conducted?

GEN ERAL COM MEN TS
In the tables 1 and 2, you should mention the values obtained even the ones that were not statistically significant.
The figures mentioned in the manuscript were not present in the reviewed data available.

Reviewers' 2 comments Response by authors Location in text
Thank you for providing your revised paper for review. Please see below my further comments: • I think currently the paper is too long and making it a little shorter could make it easier to follow. For example, maybe more emphasis on the adjusted associations in the text rather than the univariate results could help.
As requested by the reviewer, we shortened the univariate data in the results' section. The deleted section has been replaced by a supplementary table (Supplementary Table 1. Descriptive statistics of the older inpatients' health status).

Lines 227-234
• I think the definition of the unplanned nursing home admission and patients included in the analyses need more clarification. For example: How planned admissions to nursing home among the included patients could be distinguished? It is mentioned in the responses to my previous comments that patients returned home/deaths were excluded from the entire Thank you for your comment. This definition is key to understanding the manuscript. We hope it is much clearer now: "Our study defined 'unplanned nursing home admission' as the impossibility for a formerly home-dwelling older adult inpatient to return there after hospital discharge, and this included any new admission to a nursing home following an acute care admission [2]. All the patients included in the study followed home to hospital to nursing home pathway."

Population and data collection
Lines 140-144 analyses? and all patients followed a home to hospital to long-term residential care facility.
I think it is worth making this clearer in the paper including what a long-term residential care facility includes (e.g., one is nursing home which is the outcome of interest).
And what was the reason for any of the exclusions, the number excluded, and any potential biases caused by the exclusions needs to be mentioned and discussed.
To simplify, "long-term residential care facility" has been removed and replaced by "nursing home".
We completed the manuscript with the number and reasons of exclusions: "Since where patients had arrived from and where they were discharged to were two distinct variables, the dataset was recoded and customised to identify the number of older adult inpatients admitted straight from their homes and then discharged to a nursing home (n = 903) or returning to their homes (n = 13,802), as presented in a previous paper [24]. Therefore, older adults who died during hospitalisation (as assessed by the Valais Hospital's healthcare staff) were automatically excluded (n = 131)." In addition, we completed the study limitations' section as follows: "Additionally, our dataset was based on routinely collected data, and we were unable to control for potential data assessment errors made by the Valais Hospital's healthcare staff at discharge. Moreover, we were unable to assess deceased patients' death certificates as these were unavailable and beyond the scope of our study." Following the above comment, clarifications needed regarding the following sentence at line 252 of the results section: "On average, older adults whose discharge to a nursing home was unplanned had more prescribed drugs than those "returning home" [10.9 (SD = 3.9) drugs vs 8.9 (SD = 3.2)]".
We have adapted this sentence as follows: "On average, home-dwelling older adults discharged to a nursing home had more prescribed drugs than those returning to their home [10.9 (SD = 3.9) drugs vs 8.9 (SD = 3.2)].". This is mentioned in the Discussion section: "As might be expected, older adults who underwent an unplanned nursing home admission had more prescribed drugs than those returning home. Our results were in line with the Not having access to the linked data and data from other hospitals and any potential biases caused by these and by any of the exclusions need to be highlighted more clearly in the discussion section.
To address this relevant comment, the following statement was added in the Study strengths and limitations section: "The Swiss Federal Statistical Office collects minimal annual data from public and private hospitals (number of hospitalisations, ICDdiagnoses, length of stay, place of discharge, age and sex), but these indicated that our data were similar to those from other cantons with anlogous healthcare structures [43]. However, we did not have access to more detailed data with which to compare with our dataset and explore potential biases or significant differences. Nevertheless, the Valais Hospital is the third largest hospital in Switzerland with more than 1,000 beds and over 35,000 hospitalisations per year. Therefore, our findings could provide information to help better define which integrated healthcare approaches could be implemented to attenuate the risk factors associated with unplanned nursing home admission following an acute hospital admission or readmission.".

Study strengths and limitations
Lines 388-397 • I think the data analysis strategy section need some rewording to become clearer and shorter. For example: -Please replace the word bivariate in the paper with the word univariate to more clearly refer to the unadjusted analysis conducted. For example, at line 199 something like "univariate analysis using logistic regression models was conducted to investigate…" instead of "multiple bivariate logistic regression analysis was conducted using crosstabulations to investigate…".
We have adapted the manuscript to specify that bivariate analyses were unadjusted and multivariate analyses were adjusted. We have also clarified the description of the analytical part to be clearer.
After discussing this as a team, we believe that including odds ratios would be redundant alongside percentages resulting from cross-tabulations, given the fact that those first Data analysis strategy Including the odds ratios in these tables could also be useful. And it seems that variables significantly associated with unplanned nursing home admission in the univariate analysis, were used to develop the multivariable models using GEE models.
results are unadjusted. The paper does present odds ratios for the multivariate-adjusted analysis.
-Line 207: this sentence is not very clear: "This baseline model was completed using the drugs prescribed to older inpatients who underwent unplanned nursing home admission." Does it mean that the base line model was completed by adding to it drugs that were found significantly associated with unplanned nursing home admission based on the univariate analysis? Did adding drugs to the model make any changes in the associations observed and reported in the baseline model in the previous figure/supplementary table?
Thank you for your comment. We have revised our text as follows: "This adjusted baseline model was then completed by adding drugs that were found to be significantly associated with unplanned nursing home admissions in the previous analysis.". Adding these drugs did not cause any changes in Table 1.

Data analysis strategy
Lines 212-214 -How authors have ensured that their final multivariable model is not overfitted? How many parameters were included in the final multivariable model?
We completed the Data analysis strategy section as follows: "The multivariable analysis model included 52 Level 2 ATC drug classes, respecting the good practices for logistical regressions involving large population-based samples [27]."

Data analysis strategy
Lines 210-212 -As is also indicated in the paper the GEE method is used to estimate population-averaged estimates. Please delete lines 213-216 (or please modify).
As requested by the reviewer, lines 213 to 216 have been deleted.

Data analysis strategy
Lines 217-220 • Line 242: with five or more diseases We corrected as suggested: "Being concomitantly affected by several diseases increased the prevalence of unplanned nursing home admission, from 1.8% (n = 5) for

Results
Lines 248-250 older adults with a single disease (ICD-10) to 6.8% (n = 797) for those with five or more diseases." • Line 246, for investigating the association between number of drugs prescribed and unplanned nursing home admission in the univariate analysis authors could use logistic regression consistent with other univariate analyses performed.
As mentioned before, we believe that including odds ratios would be redundant alongside percentages resulting from cross-tabulations, given the fact that those first results are unadjusted. However, considering the reviewer's comment, we have added the number of prescribed drugs in Table 1.

•
In the discussion section line 324, please modify the wording to make it clear that the comparison being mentioned is related to the percentages in the unadjusted analysis, as with the current wording (i.e., tenfold higher risk) it could be confused with relative risks/odds ratios and the adjusted analyses.
Thank you for this important remark. We changed our wording as follows: "Very old inpatients (≥ 90 years old) had an almost tenfold higher risk17.5% (19.7% vs 2.2%) more chance of an unplanned nursing home admission than those aged 65-69."

Discussion
Lines 333-334 • Study strengths and limitations, line 378: authors have indicated that their findings regarding the single hospital, included in their study, could be generalised to other regions of Switzerland. Please include in the paper how authors came to this conclusion. Any representativeness analysis been conducted?
We completed the Study strengths and limitations section with the following statement: "The Swiss Federal Statistical Office collects minimal annual data from public and private hospitals (number of hospitalisations, ICDdiagnoses, length of stay, place of discharge, age and sex), but these indicated that our data were similar to those from other cantons with analogous healthcare structures [43]. However, we did not have access to more detailed data with which to compare with our dataset and explore potential biases or significant differences. Nevertheless, the Valais Hospital is the third largest hospital in Switzerland with more than 1,000 beds and over 35,000 hospitalisations per year."

Reviewers' 3 comments
Response by authors Location in text *In the tables 1 and 2, you should mention the values obtained even the ones that were not statistically significant. [NOTE FROM THE EDITORS: we agree, all p values should be reported, including non-significant ones).
As requested, we have added all the p-values to Tables 1 and 2. Table 1 and Table  2 *The figures mentioned in the manuscript were not present in the reviewed data available.
We are sorry that these figures were not visible for your review of the manuscript, although we were careful to closely follow the publisher's instructions. We will revise the resubmission so that you can see them.