Data-Mining-Based Detection of Adverse Drug Events Emmanuel - - PowerPoint PPT Presentation

▶
data mining based detection of adverse drug events
SMART_READER_LITE
LIVE PREVIEW

Data-Mining-Based Detection of Adverse Drug Events Emmanuel - - PowerPoint PPT Presentation

Aug 18, 2023 360 likes •609 views

Data-Mining-Based Detection of Adverse Drug Events Emmanuel CHAZARD, Cristian PREDA, Batrice MERLIN, Grgoire FICHEUR, the PSIP consortium, Rgis BEUSCART 2009-08-31 1 MIE Sarajevo August 2009 PSIP: a research project (7 th FP ICT)

slide-1
SLIDE 1

MIE Sarajevo – August 2009

2009-08-31 1

Data-Mining-Based Detection of Adverse Drug Events

Emmanuel CHAZARD, Cristian PREDA, Béatrice MERLIN, Grégoire FICHEUR, the PSIP consortium, Régis BEUSCART

slide-2
SLIDE 2

MIE Sarajevo – August 2009

2009-08-31 2

PSIP: a research project (7th FP – ICT)

Consortium: 13 partners

1/ Hospitals

France, Denmark With / without CPOE

2/ Industry:

Oracle, IBM, Medasys (CPOE editors) Vidal (pharmaceutical Kbase)

3/ Academic teams

Data & Semantic mining, Decision Support Systems, Human Factors Engineering

Duration: 40 months (Jan 08  April 2011)

slide-3
SLIDE 3

MIE Sarajevo – August 2009

2009-08-31 3

  • Adverse Drug events (ADEs)

– Altmost 10% of stays – Less than 2% would be declarated – Responsible of 98000 deaths each year in USA

  • Objective

– Propose new methods to prevent ADEs – Develop automated rules to detect them – Integrate rules in a CDS system generating relevant alerts to the physician

Patient Safety Through Intelligent Procedures in Medication

slide-4
SLIDE 4

MIE Sarajevo – August 2009

2009-08-31 4

Why use data-mining to detect ADEs?

  • Chart review is time-consuming
  • With data-mining, we are able to analyze 10,000

records in few minutes

  • Data-mining may overcome the inevitable limits
  • f expert knowledge : thousands ADEs

described in the litterature

  • Data-mining may detect complex and

sometimes combined ADEs that an expert may not necessarily identify

slide-5
SLIDE 5

MIE Sarajevo – August 2009

2009-08-31 5

Data-mining based rules generation

Data What we have How we get it Events

potential causes and effects

Data aggregation Statistical associations

effects linked to causes

Statistical analysis (trees…) Drug linked events = rules Bibliographic analysis Confidence of rules

In each medical department

Evaluation of rules

slide-6
SLIDE 6

MIE Sarajevo – August 2009

2009-08-31 6

Main reasoning to detect ADE

  • ADE: injury caused by medical management rather

than the underlying condition of the patient

  • The ADE event:

– Is not declared in the stay – requires a specific human case review – Is hidden in the data

  • Data-mining: three-steps procedure

– Identify a kind of traceable incident = “effect” – Automatically find a statistical association with some drugs in combination with other causes – The coincidence of causes and effects generate a rule which allow us to detect stays with ADE

slide-7
SLIDE 7

MIE Sarajevo – August 2009

2009-08-31 7

Data-mining based rules generation

Data What we have How we get it Events

potential causes and effects

Data aggregation Statistical associations

effects linked to causes

Statistical analysis (trees…) Drug linked events = rules Bibliographic analysis Confidence of rules

In each medical department

Evaluation of rules

slide-8
SLIDE 8

MIE Sarajevo – August 2009

2009-08-31 8

From available data to causes and effects

Data can be considered as:

  • Cause or context of an ADE
  • Effect which is a potential manifestation
  • f an ADE

Example on diagnosis:

  • Cause: chronic diseases, reason of the

admission

  • Effect: acute events during the stay

Example on lab results:

  • Cause: abnormality existing at admission
  • Effect: abnormality got during the stay

Example on drugs:

  • Cause: prescription of the day
  • Effect: antidotes…
slide-9
SLIDE 9

MIE Sarajevo – August 2009

2009-08-31 9

Medical Data Bases

25,000 records

  • Copenhagen Hospitals (University Hospitals, DK)

– Cardiology & internal medicine: 2,700 records

  • Rouen hospital (University Hospital, FR)

– Cardiology & internal medicine: 800 records

  • Denain hospital (General Hospital, FR)

– Surgery: 2,600 records – Gynecology obstetrics: 1,800 records – Medicine A: 1,700 records – Medicine B: 900 records

  • Lille hospital (University Hospital, FR)

– Geriatry: 15,000 records

slide-10
SLIDE 10

MIE Sarajevo – August 2009

2009-08-31 10

From available data to events

Variables have to be interpreted

Data : values of natremia min = 135 max = 145 Event : hyponatremia = 1 start on day 2 stop on day 4 hyponatremia = 0 elsewhere

mmol/l

2 4

natremia hyponatremia

1

135 125 binary days days

slide-11
SLIDE 11

MIE Sarajevo – August 2009

2009-08-31 11

From available data to events

days

1 1 2 3 4 5 6

Antiepileptic

days

Antiepileptic

binary

Example on drugs

Data : antiepileptic day 4 antiepileptic day 5 antiepileptic day 6 Event : antiepileptic = 1 start on day 4 stop on day 6 antiepileptic = 0 before day 4 4 6

slide-12
SLIDE 12

MIE Sarajevo – August 2009

2009-08-31 12

Data aggregation: overview

  • Available data:

– Complex data scheme with 7 tables, 91 fields – Potentially more than 30000 different variables – Too numerous and redundant codes

  • E.g. Diagnosis (ICD 10): 18 000 possible codes
  • E.g. Drugs (ATC) : 5 400 possible codes
  • => Need to simplify the data
  • Aggregated data:

– One flat table containing one row per stay – 576 “cause or context” variables – 55 “possible effect” variables (lab values++)

slide-13
SLIDE 13

MIE Sarajevo – August 2009

2009-08-31 13

Data-mining based rules generation

Data What we have How we get it Events

potential causes and effects

Data aggregation Statistical associations

effects linked to causes

Statistical analysis (trees,…) Drug linked events = rules Bibliographic analysis Confidence of rules

In each medical department

Evaluation of rules

slide-14
SLIDE 14

MIE Sarajevo – August 2009

2009-08-31 14

Data Mining Methods

  • General principle of our analysis:

1) investigate the causes associated with the effect 2) Is there any drug among these causes?

  • Statistical methods already used in PSIP:

– Regression Trees (CART) – Multiple Correspondence Analysis – Logistic Regression Analysis – Principal Component Analysis – Association Rules

slide-15
SLIDE 15

MIE Sarajevo – August 2009

2009-08-31 15

Appearance of a too low INR (INR < 2) Risk of Thrombosis

60% 0% No No Yes Yes No Too high INR at entry? No Yes No Yes No Yes No Yes Age > 78.5? Yes Vitamin K antagonist ? Prokinetic drug ? Beta lactam antibacterial? 1,08% 0,8% 0,5% 7,75% 66,7% 4,8% 2,65% 30% 0% 29,2% 58,3% 20% 85,7% Age > 76.25? Hypoalbuminemia?

  • Ex. of tree obtained from

the method CART (classification and regression tree)

  • We are you looking for

the variables (causes) associated with an effect

  • Can be read at each

node:

  • the name of each

variable used in the regression

  • The confidence at

this level of the regression

slide-16
SLIDE 16

MIE Sarajevo – August 2009

2009-08-31 16

60% 0% No No Yes Yes No Too high INR at entry? No Yes No Yes No Yes No Yes Age > 78.5? Yes Vitamin K antagonist ? Prokinetic drug ? Beta lactam antibacterial? 1,08% 0,8% 0,5% 7,75% 66,7% 4,8% 2,65% 30% 0% 29,2% 58,3% 20% 85,7% Age > 76.25?

Appearance of a too low INR (patient with anticoagulation) - Rule N°1

Rule enunciation: Lab(previous too high INR)=1 & MedInfo(age)>78.68 & Lab(previous hypoalbuminemia)=1 ⇒ Appearance of a too low INR Rule characteristics: Support: 6 Confidence: 86%

7 stays match the conditions, 6 of them present the effect (86%=6/7)

Outcomes: 0% death avg duration: 13.4 days

Hypoalbuminemia?

slide-17
SLIDE 17

MIE Sarajevo – August 2009

2009-08-31 17

60% 0% No No Yes Yes No No Yes No Yes No Yes No Yes Age > 78.5? Yes Beta lactam antibacterial? 1,08% 0,8% 0,5% 7,75% 66,7% 4,8% 2,65% 30% 0% 29,2% 58,3% 20% 85,7% Age > 76.25? Hypoalbuminemia?

Appearance of a too low INR (patient with anticoagulation) - Rule N°2

Rule enunciation: Lab(previous too high INR)=0 & Drug(vitamin K antagonist)=1 & Drug(prokinetic)=1 ⇒ Appearance of a too low INR Rule characteristics: Support: 4 Confidence: 67%

6 stays match the conditions, 4 of them present the effect (67%=4/6)

Outcomes: 16.67% death avg duration: 15 days

Vitamin K antagonist ? Prokinetic drug ? Too high INR at entry?

slide-18
SLIDE 18

MIE Sarajevo – August 2009

2009-08-31 18

Data-mining based rules generation

Data What we have How we get it Events

potential causes and effects

Data aggregation Statistical associations

effects linked to causes

Statistical analysis (trees,…) Drug linked events = rules Bibliographic analysis Confidence of rules

In each medical department

Evaluation of rules

slide-19
SLIDE 19

MIE Sarajevo – August 2009

2009-08-31 19

Appearance of a too low INR (patient with anticoagulation) - Rule N°1

No No

Yes

Yes No

Too high INR at entry?

No

Yes

No

Yes

No Yes No Yes

Age > 78.5?

Yes Vitamin K antagonist ? Prokinetic drug ? Beta lactam antibacterial? 1,08% 0,8% 0,5% 7,75% 66,7% 4,8% 2,65% 30% 60% 0% 0% 29,2% 58,3% 20%

85,7%

Age > 76.25?

Hypoalbuminemia?

Rule enunciation: Lab(previous too high INR)=1 & MedInfo(age)>78.68 & Lab(previous hypoalbuminemia)=1 ⇒ Appearance of a too low INR Rule characteristics: Support: 6 Confidence: 86%

7 stays match the conditions, 6 of them present the effect (86%=6/7)

Outcomes: 0% death avg duration: 13.4 days

Too high INR means Hypocoagulation (risk

  • f bleeding): INR > 5

Interpretation: When a patient is admitted for a too high INR (risk of bleeding), if age>78 and hypoalbuminemia, then a too low INR (risk of thrombosis) appears with a 86% probability.

slide-20
SLIDE 20

MIE Sarajevo – August 2009

2009-08-31 20

Appearance of a too low INR (patient with anticoagulation) - Rule N°2

No

No Yes

Yes

No

Too high INR at entry?

No Yes No Yes No Yes No Yes Age > 78.5?

Yes Vitamin K antagonist ? Prokinetic drug ?

Beta lactam antibacterial?

1,08%

0,8% 0,5% 7,75%

66,7%

4,8% 2,65% 30% 60% 0% 0% 29,2% 58,3% 20% 85,7% Age > 76.25? Hypoalbuminemia?

Rule enunciation: Lab(previous too high INR)=0 & Drug(vitamin K antagonist)=1 & Drug(prokinetic)=1 ⇒ Appearance of a too low INR Rule characteristics: Support: 4 Confidence: 67%

6 stays match the conditions, 4 of them present the effect (67%=4/6)

Outcomes: 16.67% death avg duration: 15 days

INR at entry is correct and patients have Vitamin K antagonists When they are prescribed simultaneously a prokinetic drug, the risk of a too low INR (and consequently the risk of thrombosis) is

67 %

slide-21
SLIDE 21

MIE Sarajevo – August 2009

2009-08-31 21

Data-mining based rules generation

Data What we have How we get it Events

potential causes and effects

Data aggregation Statistical associations

effects linked to causes

Statistical analysis (trees,…) Drug linked events = rules Bibliographic analysis Confidence of rules

In each medical department

Evaluation of rules

slide-22
SLIDE 22

MIE Sarajevo – August 2009

2009-08-31 22

Interest of multi-site data- mining and rules evaluation

slide-23
SLIDE 23

MIE Sarajevo – August 2009

2009-08-31 23

Conclusion

  • 242 rules of detection have already been

validated

  • We compared our rules with other available rules

databases (Vidal, David Bates' team ): several rules in common

  • 2 chart reviews in Denain and Copenhagen to

evaluate the ability of rules to detect stays with ADEs

  • Implementation of rules in CDSS in progress
slide-24
SLIDE 24

MIE Sarajevo – August 2009

2009-08-31 24

Thank you for your attention