運用“購物籃分析技術”探討滯留急診超過24小時病患特性

(1)

24

* ** *** **** ***** 24 111,514 24 18.1% 15,454 24 24 20 0~4 65

Original Articles

* ** **** ***** 95 8 21 95 12 18 96 8 7 168

(2)

24 [1] [2] 24 87 24 48 72 48 [ 3 ] W a l l e r , Hohenhaus, Shah et al. [4]

(3)

data mining 24 1 24 2 3 Data Warehouse Data Mining [5] Market Basket Analysis

association rule ex. 250 300 111,514 24 CLEAN DATA 15,454 15,454 85,330 7.481 p-value=0.0062 p-value=0.001 [6] Pre-Process SPSS 10.0 for Windows PolyAnalyst 4.0 Basket Analysis 24

(4)

recode

PolyAnalyst Market Basket Analysis PolyAnalyst 4.0 Minimum support Minimum confidence support combination A B A B A C B D confidence A B Confidence A B =Support A U B Support A Confidence B A =Support A U B Support B association rule =0 =10 20 413.17 24 18.1% 1 6 , 9 9 5 439.30 386.13 0 ~ 4 1 9 9 . 3 7 75 951.12 55 24 25% 55~64 26.7% 65~74 30.4% 75 32.6% 624.90 24 25.7% 241.91 2 1 0 . 7 5 195.05 167.97 882.22 31.8% 24 592.98 249.87 217.80 1505.46 45.0% 24

(5)

1035.24 1189.96 ߒΙȈ੾௉ᅗ੼ࡨຨਢ໢ϞࣺᜰϷݙȞWᔮۡ ᡐ౴ኵϷݙȟ ᡐ໶Ӫᆎ ᜸տ ኺҏኵ ᅗ੼຺ႆ ωਢ ܚլШ౥ ҁ֯঄ ȞϷមȟ ኿ྥ৯ W঄ʝ)঄ ੾௉឴ܒ ܒտ ظ τ ԑឭ aྑ aྑ aྑ aྑ aྑ aྑ aྑ aྑ ྑоΰ ൷ᚂ឴ܒ ऋտ ϱऋ Ѵऋ ڋऋ ౰ऋ бऋ ᔮ༌Ϸ઻ Ι઻ Π઻ έ઻ Ѳ઻ پ଱ПԒ ௿៖ٙ т଱ᙽΣ ߞຨᙽΣ Ռ՗؏Σ ׭ΣܖܲΣ ௰חܖᎈා ຝȋ S

(6)

1081.57 24 36.6% 685.33 , , 548.26 71.49 ߒΙȈ੾௉ᅗ੼ࡨຨਢ໢ϞࣺᜰϷݙȞWᔮۡ ᡐ౴ኵϷݙȟȞ៉ȟ ᡐ໶Ӫᆎ ᜸տ ኺҏኵ ᅗ੼຺ႆ ωਢ ܚլШ౥ ҁ֯঄ ȞϷមȟ ኿ྥ৯ W঄ʝ)঄ ᚔ଱ࡣଢ଼ө ՞଱ ߞຨݽᕛ ᙽ଱ ໠Φ ଠ଱ ԫι Ռଢ଼ю଱ فಛ੽੾Ϸ᜸ ༈ࢗ੾Ѕசҡᙫ੾ ရዴ ϱϷݪȂᕊᎴȂཱིങфᗂڷջ࣫੽੾ ՖశڷഅՖᏢۢ੽੾ ᆠડራᛤ ડငفಛڷཐឈᏢۢ੽੾ ඉᕗفಛ੽੾ ڳ֜فಛ੽੾ ੑϽඉᕗفಛ੽੾ ݪ׍ҡ෥فಛ੽੾ ᛄѹȂҡ౰Ѕ౰ࡣӫځ੿ ҪጳЅҪήಢᙑ੽੾ ՊՈ଼ᓫفಛЅ๖ጚಢᙑ ӑЈ౴ல ໊౰෈੽੾ ੿ঐȂ኉ঐЅຨᘞФ݂ϞӨᆍ੾ݷ ཬ༌Ѕϛࢳ ཬ༌ЅϛࢳϞѴӰ၄шϷ᜸ ኇ៪୊ஶӰશЅ୊ஶ݈୛Ϟ၄шϷ᜸ ຝȋ S

(7)

24 POLYAN-ALYST Basket Analysis A B A B A B A B 0% 10% 20 1. 0~4 0~4 ߒΠȈᅗ੼ࡨຨ຺ႆωਢ੾௉ϞᜰᖒೣࠌȞ$VVRFLDWLRQ5XOHVȟ ᜰᖒೣࠌ Лࡻ࡙ ߬ᒦ࡙ ׽๡࡙ ܲΣࡨຨȃڋऋÆaྑ aྑȃڋऋÆܲΣࡨຨ aྑȃܲΣࡨຨÆڋऋ ѴऋÆཬ༌Ѕϛࢳʝཬ༌ѴӰ၄шϷ᜸ ཬ༌Ѕϛࢳʝཬ༌ѴӰ၄шϷ᜸ÆѴऋ ౰ऋÆᛄѹЅ౰ࡣӫځ੿ ᛄѹЅ౰ࡣӫځ੿Æ౰ऋ ௿៖ٙپࡨຨʝ௰חپࡨຨÆΙ઻ Ι઻Æ௿៖ٙپࡨຨʝ௰חپࡨຨ ௰חپࡨຨʝᎈාپࡨຨÆྑоΰ ྑоΰÆ௰חپࡨຨʝᎈාپࡨຨ ໠ΦÆཬ༌Ѕϛࢳ бऋÆੑϽඉᕗفಛ੽੾ʝኇ៪୊ஶӰશ ડငفಛ੽੾Æaྑ aྑÆડငفಛ੽੾ ߞຨᙽΣʝԫιÆရዴ ရዴÆߞຨᙽΣʝԫι ϱϷݪջ࣫੽੾ʝԫιÆ௿៖ٙ ߞຨᙽΣÆӔӱຨ੾Ρ ᆠડራᛤÆaྑ

(8)

15,454 24 5.14% 86.87% 0~4 2. 0~4 0~4 15,454 24 5.14% 0~4 73.59% 3. 0~4 0~4 15,454 24 5.14% 0~4 90.64% 4. 15,454 24 6.13% 31.31% 5. 15,454 24 6.13% 84.55% 6. 15,454 24 0.36% 72.37% 7. 15,454 24 0.36% 74.32% 8. 119 119 15,454 24 5.18% 119 o r 25.44% 9. 119 119 15,454 24 5.18% 45.51% 119 or 10. 75 75 15,454 24 5.76% or 33.97% 75 11. 75 75 15,454

(9)

24 5.76% 75 36.52% or 1 2 . 15,454 24 0.41% 21.21% 13. 15,454 24 0.56% 77.68% or 14. 5~14 5~14 15,454 24 0.46% 13.73% 5~14 15. 5~14 5~14 15,454 24 0.46% 5~14 11.22% 1 6 . 15,454 24 0.90% or 22.35% 1 7 . 15,454 24 0.90% 11.20% or 18. 119 119 15,454 24 0.47% or 24.01% 119 1 9 . 15,454 24 0.36% 10.32% 72 20. 35~44 35~44 15,454 24 0.34% 19.56% 35~44 20 24

(10)

[6] 2000 413 439 386 3 1 9 332 308 80 100 1985 1. 2. 24 45% 36.3% 41% 24 24

(11)

1 2 3 Association Rules a. 0~4 90.64% b. 0~4 73.59% c. 84.55% d. 31.31% e. 74.32% f. 72.37% a b 0~4 90.64% 0~4 73.59% 0~4 0~4 c d e f 24 g. 119 h. i. j. 119 k. g h i j k 24

(12)

l. 75 m. 55~64 65~74 n. 65~74 l m n 24 o. p. q. 5~14 r. 15~24 s. 35~44 24 24 [7] 48 24 24 55

(13)

18.1% 24 1 2 3 4 24

(14)

1. 2002 2. 1984 3. 2001

4. Waller AE, Hohenhaus SM, Shah PJ, Stern EA. Development and validation of an emergency department screening

and referral protocol for victims of domestic violence. Ann Emerg Med 1996; 27: 754-60. 5. OLAP 2002 9-12 6. 2 0 0 0 4 8 2001 6 4 235-245

(15)

Apply Basket Analysis to Explore Characteristic of

Patients Stayed Emergency Department Over

24 Hours

Hsin-Kai Chou, I-Chiu Chang, Hsin-Ginn Hwang, Ming-Tsu Tsai, Lin-Chung Woung

Abstract

Objectives: Traditionally, the algorism of basket analysis in Data Mining

is often used for business marketing, and the combination of products which are purchased together will be explored by large amounts of transaction data; however, this study also applied it to analyzed the characteristic of patients stayed at the emergency department over 24 hours.

Methods: 15,454 patients stayed at the emergency department over 24

hours in one medical center were screened from total 85,330 emergency patients in one year duration, and the logistic regression and basket analysis a Data Mining tool were used to analyze attributes of patient such as age, degree of triage, medical specifics, the way of coming, the way of leaving and the disease classification.

Results: The results of logistic regression analysis had indicated that the

attributes of gender, age, degree of triage, the way of leaving and health expense significantly influenced the status patient stayed over 24 hours or not, and then the basket analysis also produced 20 association items and rules.

Conclusion: We found some specific group needed to be managed (for

example, child patient during 0 to 4 years old, pregnancy or postpartum complication, elder upper 65 years old suffered from specific disease such as circulatory system disease, endocrine immune disease, congenital abnormal disease). Otherwise, the Basket Analysis also display something abnormal characteristics which were rarely found before and then suggested for further Hsin-Ginn Hwang, 168, University Rd., Min-Hsiung Chia-Yi, Taiwan

Received: August 21, 2006 Revised: December 18, 2006 Accepted: August 7, 2007

(16)