Transcript Development and Validation of a Prognostic Model for

02/15
Development and Validation
of a Prognostic Model for
predicting Adverse Drug
Reactions in Children
HERGIBO F., MIMOUNI Y., LAJOINIE A., CASTELLAN A-C., KASSAI B., NGUYEN K.A.
INSERM CIC 201, EPICIME, Laboratoire de Biométrie et Biologie Evolutive UMR5558
CNRS-Université Claude Bernard Lyon 1, Hospices Civils De Lyon, Lyon, France
Introduction

Adverse Drug Reactions (ADRs) detected by spontaneous report are
underreported

Others methods


Chart review

Cooperation between the pharmacovigilance team and clinicians

Direct observation

“Trigger tools”: data element within health records that may identify or
predict an AE (clinical, medication or laboratory data) associated with chart
review
Prognostic model: strategy to predict and prevent ADRs

2 studies in adults (Bates 1999, Evans 2005)

None in children
Bates et al. 1999

Methods


Nested case-control study within a cohort

4108 admissions

11 units (stratification, randomization)

2 hospitals

6 months
2 levels of analyses:

limited set of variables available for all patients using computerized
data from 1 hospital

larger set of variables for the case patients (with an ADE) and matched
controls (same unit with the most similar pre-event length of stay) from
both hospitals
Outcomes: presence of an ADE, preventable ADE or severe ADE
Results
Independent Predictor

Cohort Analysis
All ADEs
Preventable ADEs

Case-control Analysis
OR (CI95%)
Electrolyte concentrate
1.7 (1.1 - 2.5)
Diuretic
1.7 (1.0 - 2.6)
Medical ward admittance
1.6 (1.1 - 2.3)
Platelet category
4.5 (1.6 - 12.9)
Antidepressant
3.3 (1.3 - 7.9)
Antihypertensive agent
2.9 (1.4 - 4.4)
Medical ward admittance
2.2 (1.1 - 4.4)
Electrolyte concentrate
2.1 (1.1 - 4.1)
Independent Predictor
OR (CI95%)
All ADEs
Exposure to psychoactive drugs
2.1 (1.3 - 3.6)
Severe ADEs
Cardiovascular drugs
2.4 (1.3 – 4.5)

Small number of patients

Little power of predictors

No generalizability to other care settings: variation of
data from site to site

No specific patient groups and type of events

Risk stratification approach unlikely productive
Limitations
Evans et al. 2005
Methods

Conditional logistic regression

Analysis of ADEs by therapeutic class of drugs and severity

10 year

1 hospital

Case matched with up to 16 control patients
Results
Risk Factor
Patients
Characteristics

4376 ADEs
Drug
Administration
Patient Type
OR (CI95%)
Female
1.5 – 1.7
Age
0.7 – 0.9
Weight
1.2 – 1.4
Creatinine Clearance
0.8 – 4.7
Number of Comorbidities
1.1 – 12.6
Dosage
1.2 – 3.7
Administration Route
1.4 – 149.9
Number of Concomitant drugs
1.2 – 2.4
Service
1.2 – 4.9
Nursing Division
1.5 - 3.8
Diagnosis-related Group
1.5 – 5.7

The computer-prompted method of surveillance of ADEs may have
systematically missed certain categories of ADEs

Limitations
Antineoplastic agents and anesthesia probably underreported

No access to nurse staffing levels: no increased patient census,
patient turnover, and nursing acuity as risk factors

Liver disease was not identified as a significant risk factor for ADEs;
this may be due to misclassification
Objectives

Primary objective:
To develop and validate the EREMI trigger tool prognostic model
for predicting of ADRs in a hospitalised paediatric population

Secondary objective:
To describe the ADRs detected by the model (by age groups,
medications the most incriminated, the most frequent ADRs)
Method
Study Design

Database study (EREMI data)

EREMI: observational, multicenter, prospective study which assesses the
relationship between ADRs and unlicensed/off-label drug use in
hospitalized children
Study
Population
EREMI Patients
Inclusion Criteria
Children 0-15 years

EREMI patients enrolled
from September 2013 to
December 2014 in Lyon
> 3 days hospitalisation
> 1 drug administration
Wards:
Nephrology
Exclusion Criteria
Patients hospitalized for
an ADR
Patients
undergoing
voluntary
drug
detoxification
Psychopathology

+ patients > 15 years or
hospitalized < 3 days
Endocrinology
Patients aged 15 or more
Pneumology

high
accidents,
suicide
Neurology
Hepato-gastrology
Rheumatology
Pediatric
reanimation
death
rate:
violence,
Patients who did not take
any drugs during their
hospital stay
Ethics - EREMI

Funded by the ANSM

Favourable opinion of the CCTIRS on 03/10/2013

Subjected to an authorization application from the CNIL

Consent not required: information sheet sufficient to meet regulatory
requirements

CPP submission is on going
Electronic chart review
process for ADR
detection and
validation
Medical records
Prescriptions
● Medical history
● Prescribed drugs
● Cause of hospital admission
● Administered drugs
● Clinical observations
● Hospital discharge report
Active
Centralized
● Dosing regimen
● Drug-Drug Interactions
Detection
Health exam & lab test
results
● Physiological functions
● Biological examinations
● Evolution of the results
Trigger
Tool
Original Trigger Tool
Data Flow
Licensed/UL/OL Classification
Theriaque
EREMI
Hospital
Database
Centralised
Age
Coded
Hospital stay
Database
Administered drugs
Height
Weight
Lab results
ADR:
Active detection
Diagnoses
etc.
Nominal data
ADR:
spontaneous
reports
ADR Detection & validation
Pharmacovigilance
Regional
Centers
Pharmacovigilance
Independent
Board
Cross validation
Summary of Triggers
Variable
Trigger
Age
Weight
Demographic variables
Height
BMI
Underweight, Overweight
Creatinine plasma levels
Hypercreatininemia,
Renal
failure,
toxicity
Administered drugs
Overdosage (antidotes), hypoglycemia
Routes of administration
or hyperglycemia (inulin), hemorrhage
Doses
or
Diagnoses
(antiemetics),
blood
clot
(heparin),
vomiting
allergic
reaction
(antihistaminics), withdrawal syndrome
Variable
Trigger
Variable
Trigger
Temperature
Hypothermia, Hyperthermia, Fever
Calcium
Hypocalcemia, Hypercalcemia
Blood Pressure
Hypotension, Hypertension
Drug Dosage
Toxicity
Oxygen Saturation
Hypoxia
Chlorate
Anemia
Phosphate
Leucopenia, hyperleucocytosis
Anionic gap
Metabolic acidosis
Neutropenia, neutrophilic
Serum Glucose
Hypoglycemia, Hyperglycemia
hyperleucocytosis
Urea
Renal failure/toxicity
Eosinophils
Hypereosinophilia
Uric acid
Platelets
Thrombocytopenia
ASAT
Total IgE
Allergic reaction
ALAT
Liver failure/toxicity
Phosphatase alcaline
Hyperbilirubinemia
Gamma glutamyl transferase
Increased transaminases
Hemoglobin
Hematocrit
Leucocytes
Neutrophils
Partial Thromboplastin Time
INR
Hemorrhage
Anti-Xa
Magnesium
Hypomagnesemia,
Hypermagnesemia
Hypochloremia, Hyperchloremia
Hypophosphoremia,
Hyperphosphoremia
Total bilirubin
D-dimer
Blood clot
Triglycerides
Sodium
Hyponatremia, Hypernatremia
LDH
Potassium
Hypokalemia, Hyperkalemia
Lipase
Hyperlipidemia
Selection of
Variables
Training
Set
Multivariate Analysis
p < 0.10
Intermediate
Model
2/3 patients
Univariate Analysis
p < 0.20
Final
Model
Discrimination
Calibration
Validation
Set
1/3 patients
Conclusion

Define pronostic value of physiological and biological triggers in
hospitalized children

Further step: external validation with EREMI patients from Paris

Contribute to the development of an automated trigger tool
Thank you
TRAINING SET
2/3 of the patients
Selection of variables/triggers
- Demographic
- Medication
Prognostic
Model Statistical
Analysis
- Laboratory
Method for accounting missing values
Univariable analysis p<0.20
VALIDATION SET
1/3 of the patients
Discrimination
- Sensitivity, Specificity, predictive value
Multivariable analysis p<0.10
- C-statistic (ROC)
Backward stepwise regression
Calibration
Final multivariate model of triggers
Risk score