Patient Care Bundles

Patients diagnosed as mild cognitive impairment or dementia at memory clinic from October 2015 to April 2017 in Changhua Christian Hospital were enrolled. The clinical trial was approved by the Institutional Review Board of Changhua Christian Hospital (CCH IRB 160165). Because the design of the present study was a retrospective chart review, informed consent was waived by the Institutional Review Board of Changhua Christian Hospital. All data were recorded in the electronic medical chart with the highest confidentiality and compliance with the Declaration of Helsinki. Six hundred and eight patients who received team-approached assessment were screened, and after excluding those with incomplete data and those with a dementia subtype including fewer than 10 registered patients, the remaining 508 were included for analysis. The demographic information including gender, age, type of dementia, and clinical dementia rating (CDR) is summarized in Table 2. A majority of the patients were women (66%) with Alzheimer’s disease (72%) aged 75 years and older (82%) who had mild dementia (51%). National Institute on Aging-Alzheimer’s Association (NIA-AA),22 (link),23 (link) International Society for Vascular Behavioral and Cognitive disorders (VASCOG)24 (link) and Movement Disorder Society-Task force criteria were used for diagnosis of AD, vascular dementia and Parkinson's disease dementia, respectively.25 (link) Patients fit both possible AD by NIA-AA criteria and possible major vascular cognitive disorder by VASCOG criteria were classified as mixed dementia. The 15 care needs listed in Table 1 were then assessed.Table 2

Information Of The Patients With Dementia

Variables		Frequency	Percentage	Data Type
Gender	Male	175	34	M
Female	333	66	F
Age	Less than 65 years old	17	3	0
	65–69 years old	20	4	1
	70–74 years old	56	11	2
	75–79 years old	147	29	3
	80–84 years old	145	29	4
85 years old and above	123	24	5
Type of dementia	Alzheimer’s disease	363	72	AD
	Parkinson’s disease	26	5	PDD
	Vascular dementia	93	18	VaD
Mixed dementia	26	5	MD
CDR	Very mild dementia	104	21	0
	Mild dementia	260	51	1
	Moderate dementia	89	18	2
Severe dementia	55	11	3

Abbreviation: CDR, clinical dementia rating.

The purpose of this study was to identify whether specific combinations of several care needs could be applied to particular groups of PLWD and their caregivers and thereby to provide more efficient and holistic care. That is, we intended to group the care needs for patients with dementia by gender, age, type of dementia, and dementia severity. The Apriori algorithm has been proven to be a very useful approach to discover previously unknown interesting relationships in data sets by finding rules and associations between any of the attributes by establishing support, confidence, and lift, which are defined as follows.21 ,26 ,27 (link) The support for an association rule A ⇒ B is assessed by calculating the percentage of transactions in the database containing both A and B:
(1)\documentclass[12pt]{minimal}
\usepackage{wasysym}
\usepackage[substack]{amsmath}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage[mathscr]{eucal}
\usepackage{mathrsfs}
\DeclareFontFamily{T1}{linotext}{}
\DeclareFontShape{T1}{linotext}{m}{n} {linotext }{}
\DeclareSymbolFont{linotext}{T1}{linotext}{m}{n}
\DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext}
\begin{document}
$${\rm{Support = P}}\left({{\rm{A}} \cap {\rm{B}}} \right)\break{\rm{ = }}{{{\rm{number}}\,{\rm{of}}\,{\rm{transactions}}\,{\rm{containing}}\,{\rm{both}}\,{\rm{A}}\,{\rm{and}}\,{\rm{B}}} \over {{\rm{total}}\,{\rm{number}}\,{\rm{of}}\,{\rm{transactions}}\,}}$$
\end{document}The confidence of the association rule A ⇒ B is assessed by evaluating the accuracy of the rule based on the calculation of the percentage of transactions in the database containing A and also containing B simultaneously:
(2)\documentclass[12pt]{minimal}
\usepackage{wasysym}
\usepackage[substack]{amsmath}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage[mathscr]{eucal}
\usepackage{mathrsfs}
\DeclareFontFamily{T1}{linotext}{}
\DeclareFontShape{T1}{linotext}{m}{n} {linotext }{}
\DeclareSymbolFont{linotext}{T1}{linotext}{m}{n}
\DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext}
\begin{document}
$${\rm{Confidence = P}}\left({{\rm{B}}\left| {\rm{A}} \right.} \right){\rm{ = }}{{{\rm{P}}\left({{\rm{A}} \cap {\rm{B}}} \right)} \over {{\rm{P}}\left({\rm{A}} \right)}}\break{\rm{ = }}{{{\rm{number}}\,{\rm{of}}\,{\rm{transactions}}\,{\rm{containing}}\,{\rm{both}}\,{\rm{A}}\,{\rm{and}}\,{\rm{B}}} \over {\,{\rm{number}}\,{\rm{of}}\,{\rm{transactions}}\,\,{\rm{containing}}\,{\rm{A}}}}$$
\end{document}Lift is a simple correlation measuring whether A and B are independent or dependent and correlated events as shown in Equation (3). If a rule has a lift of one, A and B are independent and no rule will be generated containing either event. If a rule has a lift greater than one, A and B are dependent and correlated positively. In practice, analysts tend to prefer rules with either high support or high confidence and usually both.28 In fact, strong rules will be found when certain minimum support and confidence conditions have been met.
(3)\documentclass[12pt]{minimal}
\usepackage{wasysym}
\usepackage[substack]{amsmath}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage[mathscr]{eucal}
\usepackage{mathrsfs}
\DeclareFontFamily{T1}{linotext}{}
\DeclareFontShape{T1}{linotext}{m}{n} {linotext }{}
\DeclareSymbolFont{linotext}{T1}{linotext}{m}{n}
\DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext}
\begin{document}
$${\rm{Lift }}\left({{\rm{A,B}}} \right){\rm{ = }}{{{\rm{P}}\left({{\rm{A}} \cap {\rm{B}}} \right)} \over {{\rm{P}}\left({\rm{A}} \right){\rm{P}}\left({\rm{B}} \right)}}$$
\end{document}The Apriori algorithm in IBM SPSS Modeler 18 was used in this study. The notations of data type defined by numerical values or abbreviations are depicted in Table 2. The input variables for antecedents were gender, age, type of dementia, and CDR. The 15 care needs developed by our hospital were classified as both antecedents and consequents. The notations for each care need were 1 or 0. If a particular care need was applied to a patient, a value of 1 was assigned. If not, a value of 0 was used. This allowed for the identification of combinations of care needs. Minimum support, minimum confidence, and lift were set to 10%, 90%, and greater than one, respectively. The rules with higher support values indicated that the care need bundles could fulfill the majority of the patients’ needs. In contrast, the rules with relatively lower support values indicated that the care need bundles may only be applicable to a small portion of the patients, ie, special needs. The settings of antecedents and consequents in the Apriori algorithm allow the decision-maker to identify what combinations of care needs would be needed according to the specific demographic information of the patients.

Patients diagnosed with mild cognitive impairment or dementia at the memory clinic of Changhua Christian Hospital from November 2015 to June 2019 were enrolled. A total of 389 male patients with VCI were identified. International Society for Vascular Behavioral and Cognitive disorders (VASCOG) criteria were used to diagnose VCI.26 (link) Patients who met the criteria for both possible AD according to the National Institute on Aging-Alzheimer’s Association (NIA-AA)27 (link),28 (link) and possible major vascular cognitive disorders according to the VASCOG were classified as having mixed dementia, and were excluded from the study. The clinical trial was approved by the Institutional Review Board of Changhua Christian Hospital (CCH IRB 160165). The need for informed consent was waived by the Institutional Review Board of Changhua Christian Hospital because the design of the study was a retrospective data analysis. All data including assessment results (such as ADL score for Care 2, Zarit’s caregiver burden score for Care 13), and team-selected care need numbers (as listed in Table 1) were recorded in electronic medical charts with the highest confidentiality and in compliance with the Declaration of Helsinki. All data needed in the present study were extracted by our information technology department after deleting any personalized information. The demographic information of the male patients with VCI including age and clinical dementia rating (CDR) score were recorded (Table 2). A majority of the included patients were aged 75 years and older (75.3%) and had mild dementia (CDR=1, 39.3%). Subjects with a CDR score of 3 were further classified as having severe dementia and extremely severe dementia. In this study, extremely severe dementia indicated nonresponsive to external stimuli and unable to communicate, and these patients met the criteria for hospice care provided by the Taiwan National Health Insurance program.Table 2

Information of the Male Patients with VCI

Variables		Frequency	Percentage	Data Type
Age	Less than 65 years old	35	8.9	1
	65–69 years old	25	6.4	2
	70–74 years old	36	9.3	3
	75–79 years old	66	17.0	4
	80–84 years old	107	27.5	5
	85 years old and above	120	30.8	6
CDR	0.5 (MCI or very mild dementia)	147	37.8	0
	1 (Mild dementia)	153	39.3	1
	2 (Moderate dementia)	61	15.7	2
	3 (Severe dementia)	27	6.9	3
	3 (Extremely severe dementia)¥	1	0.3	4

Note:^¥Extremely severe dementia: subjects are unable to communicate and nonresponsive to external stimuli.

Abbreviations: CDR, clinical dementia rating; MCI, mild cognitive impairment.

The aim of this study was to identify care need combinations for male patients with VCI and their caregivers in order to provide holistic care since patients with different ages and dementia severity may need different combinations of care needs. The Apriori algorithm was used to identify statistical correlations among attributes by setting up support, confidence, and lift.17 (link),25 (link) Jhang et al17 (link) and Lin et al25 (link) previously reported that the Apriori algorithm was a very effective approach to identify care bundles for patients with dementia and their caregivers. Therefore, we used the Apriori algorithm in this study. The descriptions of support, confidence, and lift are as follows.17 (link),25 (link),29 (link) The support of A ⇒ B is calculated by the percentage of transactions consisting of both A and B in the database expressed by Equation (1):
(1)\documentclass[12pt]{minimal}
\usepackage{wasysym}
\usepackage[substack]{amsmath}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage[mathscr]{eucal}
\usepackage{mathrsfs}
\DeclareFontFamily{T1}{linotext}{}
\DeclareFontShape{T1}{linotext}{m}{n} {linotext }{}
\DeclareSymbolFont{linotext}{T1}{linotext}{m}{n}
\DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext}
\begin{document}
$${\rm{Support}}\; = \;P\left({A \cap B} \right)\; = \;{{{\rm{number\,of\,transactions\,containing\,both\,A\,and\,B}}} \over {{\rm{total\,number\,of\,transactions}}}}$$
\end{document}The confidence of A ⇒ B is to compute the accuracy of the rule based on the proportion of transactions in the database containing A and also containing B simultaneously depicted in Equation (2):
\documentclass[12pt]{minimal}
\usepackage{wasysym}
\usepackage[substack]{amsmath}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage[mathscr]{eucal}
\usepackage{mathrsfs}
\DeclareFontFamily{T1}{linotext}{}
\DeclareFontShape{T1}{linotext}{m}{n} {linotext }{}
\DeclareSymbolFont{linotext}{T1}{linotext}{m}{n}
\DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext}
\begin{document}
$${\rm{Confidence}}\;{\rm{ = }}\;P\left({B\left| A\right.} \right)\; = \;{{P\left({A \cap B} \right)} \over {P\left(A \right)}}$$
\end{document}(2)\documentclass[12pt]{minimal}
\usepackage{wasysym}
\usepackage[substack]{amsmath}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage[mathscr]{eucal}
\usepackage{mathrsfs}
\DeclareFontFamily{T1}{linotext}{}
\DeclareFontShape{T1}{linotext}{m}{n} {linotext }{}
\DeclareSymbolFont{linotext}{T1}{linotext}{m}{n}
\DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext}
\begin{document}
$$ = {{{\rm{number\,of\,transactions\,containing\,both\,A\,and\,B}}} \over {{\rm{number\,of\,transactions\,containing\,A}}}}$$
\end{document}Lift in Equation (3) is to measure the correlation between A and B. If a lift is one, A and B are statistically independent indicating no rule will be generated containing either event. In contrast, if a lift is larger than one, A and B are dependent and positive correlated. In practice, analysts tend to prefer rules with either high support or high confidence, and usually both.30
(3)\documentclass[12pt]{minimal}
\usepackage{wasysym}
\usepackage[substack]{amsmath}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage[mathscr]{eucal}
\usepackage{mathrsfs}
\DeclareFontFamily{T1}{linotext}{}
\DeclareFontShape{T1}{linotext}{m}{n} {linotext }{}
\DeclareSymbolFont{linotext}{T1}{linotext}{m}{n}
\DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext}
\begin{document}
$${\rm{Lift}}\left({A,\;B} \right)\; = \;{{P\left({A \cap B} \right)} \over {P\left(A \right)P(B)}}$$
\end{document}The Apriori algorithm in IBM SPSS Modeler 14.1 was used, and the notations of data type in age and CDR are depicted in Table 2. The input variables for antecedents were age and CDR. The 15 care needs described in Table 1 and developed by the dementia collaborative care team at Changhua Christian Hospital were classified as both antecedents and consequents. In doing so, the analyst could identify whether there was any association between the care need combinations and age and CDR of the male patients with VCI. The notations for each care need adopted a binary data set, i.e., 0 and 1. If a particular care need was applied to a patient, a value of 1 was assigned. If not, a value of 0 was used. Minimum support, minimum confidence, and lift were set to 10%, 90%, and greater than one, respectively. The rules with higher support values were care bundles that could fulfill the majority of the patients’ needs. On the other hand, the rules with relatively lower support values were the care bundles that may only be applicable to a small portion of the patients, i.e., special needs.17 (link),25 (link)

Search results were assessed in a two-stage process. Records were imported to EndNote (V.20.1, Clarivate Analytics, Philadelphia, USA) and, after duplicate removal, imported to Rayyan.35 (link) Two authors (NP and ERB) independently screened titles and abstracts using Rayyan’s blinding option. Additionally, we conducted a hand search of reference lists of all included studies and relevant reviews identified in the screening to find additional literature. Two authors (NP and ERB) independently assessed the full texts of the remaining records. Disagreements between authors were solved through discussions. Reasons for exclusion were documented (online supplemental table S2).
Inclusion criteria were as follows: (1) participants were ≥18 years; (2) more than 50% of the patients received ICU treatment; (3) an ICU care bundle (≥3 bundled measures) was compared with standard care; (4) patient-relevant outcomes were measured at ICU discharge or later; (5) original research article; (6) published in English, German or Spanish. Exclusion criteria were as follows: (1) paediatric patients; (2) no measurement of patient-relevant outcomes at ICU discharge or later; (3) records were based on expert opinion or secondary research only.