Personalization of SurgeriesThe tool for minimizing the number of readmissions
to hospitals and patient safety
Zbigniew W. RasCollege of Computing and Informatics, UNC-Charlotte
& James Studnicki
College of Health and Human Services, UNC-Charlotte
22
1. Failures of care delivery2. Failures of care coordination3. Overtreatment4. Administrative complexity5. Pricing failures6. Fraud and abuse7. Readmissions & - our area Patient Safety of research
1Berwick DM, Hackbarth AD. Eliminating waste in US health care. JAMA. 2012;307(14):E1-E4.2Lowe TJ, Partovian C, Kroch E, Martin J, Bankowitz R. Measuring cardiac waste: the Premier Cardiac Waste Measures. Am J Med Qual. 2013 May 29, DOI: 10.1177/1062860613487923.
Primary Areas of Healthcare Waste Research1
PREMIER primary area of research2
3
We focus on:
- decreasing the number of readmissions to hospitals caused by side effects
- improving patient safety
Expected outcome:
- significant reduction of the cost
4
Florida State Inpatient Databases (SID), part of the Healthcare Cost and Utilization
Project (HCUP)
• A total of over 2.5 million visit discharges from 1.5 million patients
• Patients are diagnosed with a maximum of 31 diagnostic codes.
• Demographic data: age, gender, race,……..
5
Patient has 7 diagnosticcodes discovered by doctor
Where is the problem?
Surgery(i) is recommended because ofthese 3 (marked) diagnostic codes
Surgery(i)Surgery successful??
Patient readmitted to the hospital because of the side effects caused by surgery
New diagnostic codes
15 readmissions – worst caseAverage – between 5 and 6
66
All patients in SID databasehaving d1,d2,d3 diagnostic codes assigned to them.
Where is the problem?
Surgery(i) is recommended because ofd1,d2,d3 diagnostic codes
Surgery(i)
d1,d2,d3
{d1,d2,d3} -> empty set
{d1,d2,d3} -> {d1,d3}
{d1,d2,d3} -> {d1,d3,d5}
{d1,d2,d3} -> {d1, d5,d6}
{d1,d2,d3} -> {d5,d7}
Personalized Surgeries(up to 300 per surgery)
We partitioned all patients having Surgery(i)Into clusters: 2 patients are in the same clusterif they confirm the same rule.
Rules
777
All patients in SID databasehaving all 7 diagnostic codes assigned to them.
Where is the problem?
Surgery(i) is recommended because ofd1,d2,d3 diagnostic codes
Surgery(i)
d1,d2,d3d4,d5,d6,d7
{d1,d2,d3} -> empty set
{d1,d2,d3} -> {d1,d3}
{d1,d2,d3} -> {d1,d3,s5}
{d1,d2,d3} -> {d1, s5,s6}
{d1,d2,d3} -> {s5,s7}
Personalized Surgeries(up to 20 per surgery)
s5 – listed in ALL personalized surgeriesWe should take care of s5 during Surgery(i)To take care of s5 may require additional personalized surgery which may cause new side effects
Much smaller DB
8
--------------------------------------------------------------------------------------------------------------Patient ID Visit-Number Main-Procedure-Code Set-of-Diagnoses-Codes--------------------------------------------------------------------------------------------------------------1 1 11 {11, 20, 234}1 2 44 {11, 234}1 3 98 {22,
234}--------------------------------------------------------------------------------------------------------------2 1 44 {11, 34, 99}2 2 122 {4; 34}--------------------------------------------------------------------------------------------------------------3 1 44 {11, 20, 44, 101} 3 2 92 {20, 44, 4}3 3 122 {4, 34}--------------------------------------------------------------------------------------------------------------4 1 11 {11, 22, 89}4 2 44 {11, 234, 89}4 3 122 {11, 22,
89}--------------------------------------------------------------------------------------------------------------
Red color – Negative side effect ; Green - Neutral ; Black color - positive
99 99
All patients in SID databasehaving all 7 diagnostic codes assigned to them.
Assume that d4 – hyperthyroid and without d4 personalized surgeries are much safer and haveless number of side effects. It means d4 has tobe fixed first.
Quite interesting?
Surgery(i) is recommended because ofd1,d2,d3 diagnostic codes
Surgery(i)
d1,d2,d3d4,d5,d6,d7
Surgery unsuccessful
{d1,d2,d3} -> {s5,s6}
{d1,d2,d3} -> {d1,d3,s5}
{d1,d2,d3} -> {d1, s5,s6}
{d1,d2,d3} -> {s5,s7}
Personalized Surgeries(up to 20 per surgery)
d4
1010101010
All patients in SID databasehaving all 7 diagnostic codes assigned to them.
Assume that d8 – high blood pressure and with d8 added to {d1,d2,d3,d4,d5,d6,d7} personalized surgeries are much safer and have less number of side effects. It means d8 has to be invoked before Surgery(i) takes place.
Quite surprising?
Surgery(i) is recommended because ofd1,d2,d3 diagnostic codes
Surgery(i)
d1,d2,d3d4,d5,d6,d7
Surgery unsuccessful
{d1,d2,d3} -> {s5,s6}
{d1,d2,d3} -> {d1,d3,s5}
{d1,d2,d3} -> {d1, s5,s6}
{d1,d2,d3} -> {s5,s7}
Personalized Surgeries(up to 20 per surgery)
d4 d8
111111111111
All patients in SID databasehaving all 7 diagnostic codes assigned to them.
Assume that d8 – high blood pressure and with d8 added to {d1,d2,d3,d4,d5,d6,d7} personalized surgeries are much safer and have less number of side effects. It means d8 has to be invoked before Surgery(i) takes place.Maybe d4 should fixed first?
What we have done wrong?
Surgery(i) is recommended because ofd1,d2,d3 diagnostic codes
Surgery(i)
d1,d2,d3d4,d5,d6,d7
Surgery unsuccessful
d4 d8
121212
Surgery(i)Surgery(i3)
Surgery(i2) Surgery(i6)
Passed Away
50
500
200
10
10
30
20
D1
D2
Split 500 patients using all stable features (not comorbid conditions) using minimal average entropy approach. Repeat this step for all surgeries.The trees defining different levels of personalization will differ but the same attributes will be used on the lowest level of these trees.
Use J48 (WEKA) on all stable attributes to split
Use J48 (WEKA) on all stable attributes to split
1313
Surgery(i)
Surgery(i5)
Surgery(i4)
Surgery(i3)
Surgery(i2)
Surgery(i8)
Surgery(i7)
Surgery(i6)
Passed Away
Passed Away
50
500
50
200
10
10
30
150
50
20
150
20
Passed Away
10
Probability of thepath: 2/5 * 3/4
cost, length of stay, …, side effects
D1
D2
D3
D4D5
Assume that Surgery(i), Surgery(i2), Surgery(i6) is the optimal path
1414
Surgery(i)Surgery(i3)
Surgery(i2)
Surgery(i7)
Surgery(i6)
50
500200
10
30
150
20
D1
D2
We would like to move 200 patients following the path Surgery(i), D2 to Surgery(i), D1.
P1 P2 P3 P4 D
pat1 X X X D2
pat2 X X X D2
pat3 X X X D2
pat200 X X X D2
pat201 X X D1
pat250 X X X D1
Diagnostic Codes(Diagnoses)
p1.p2.p3 -> p1.p2p1.p3.p4 -> p1.p2
Placing patients on a different path
Questions ?
15