Surveillance of gastroenteritis using drug sales data in France Mathilde Pivette, PharmD, MPH [email protected] Pr Avner Bar-Hen Dr Pascal Crépey.

Surveillance of gastroenteritis using drug sales data in

France

Mathilde Pivette, PharmD, [email protected]

Pr Avner Bar-HenDr Pascal CrépeyDr Judith Mueller

EHESP

Young Researcher Forum, Brussels, 13th November 2013

1

ContextDrug sales oNon-specific surveillance data oOutbreak detection oInfectious disease surveillance

Gastroenteritis o High frequency diseaseo ~ 3 millions GP consultations o 50 000 hospitalizations < 5 years old

EHESP

2

Objective

To assess the value of drug sales data as an early epidemic detection tool for gastroenteritis in France

oBy assessing correlation with reference data oBy determining if drug data could provide an early signal of seasonal outbreakoBy assessing prospective outbreak detection

EHESP

3

Data Stratified sample of pharmacies

•1647 in 2009 to 4627 pharmacies in 2013 (20%)

•Number of boxes sold of all products

•Prescribed/ Non-prescribed

•Data obtained at D+1

•Geographic location of the pharmacies (region)

EHESP

4

Indicator drug selection Intestinal antiinfectives antidiarrhoeals (A07A)

Intestinal adsorbents antidiarrhoeals (A07B)

Antidiarrheal microorganisms (A07F)

Other antidiarrheals (A07X)

Motility inhibitors (A07H)

Antiemetics and antinauseants (A04A9)

Oral rehydration solutions

Dietetic products for diarrhea and vomiting

EHESP

5

Selection of 8 groups (256 products)

Reference data

• Sentinel network of 1300 GP throughout France (www.sentiweb.fr)

• Acute diarrhea cases reported each week

EHESP

6

Sales of drugs for gastroenteritis and number of reported cases (Sentinel network), 2009-2012, France

Results

Prescribed drugs / cases

Non prescribed drugs/ cases

Coefficient correlation r 0,89 0,77

Time lag (week) 0 -1

Cross-correlation

EHESP

7

Epidemics detectiono Detection Method : Serfling method

Epidemic periods

Periodic baseline level

Upper limit of the CI : threshold

o Evaluation :• Detection window : Start of epidemic from Sentinel network +/- 4 weeks • Evaluation criteria:

• Sensitivity • False alert rate• Timeliness

Selection of model parameters that optimize the 3 criteria Selection of model parameters that optimize the 3 criteria

EHESP

8

The selected detection model for non-prescribed drugs allows the detection of seasonal outbreaks 2.25 weeks earlier

Detection performance of the selected model (IC 95%, cut-off 30%)Sensitivity : 100%False alert rate : 0%Mean timeliness: -2.25 weeks (min -3; median -2.5, max -1)

Detection week (Drugs)

EHESP

9

The selected detection model for prescribed drugs allows the detection of seasonal outbreaks 0.2 weeks earlier

Detection performance of the selected model (IC 99%, cut-off 30%)Sensitivity : 100%False alert rate : 0%Mean timeliness: -0.2 weeks (min -2; median 0, max +1)

Drug sales

EHESP

10

Prospective detection during 2012-2013Detection of epidemic 3 weeks earlier than sentinel network in

2012-2013

Non-prescribed Drug sales

Threshold Detection week (Drugs)

Training period

11

EH

ES

P

Example of the 2012/2013 seasonal epidemic.

First epidemic week from drug sales

First epidemic week from Sentinel network

Detection from non-prescribed drugs 3 weeks earlier than detection from reference data, with a beginning at the east of France.

Next step : regional analyses

EHESP

Discussion

12

Confirmation of the potential of drug sales analysis for gastroenteritis surveillance

o Prescribed drugs: high correlation with reported cases / No benefit for early detection

o Adequacy between the 2 sources

o Non prescribed drugs :Detection on average 2,25 weeks earlier (daily analysis: 16.7 days earlier, detection after 7 epidemics days)

o Purchase of drugs during the early phase of illnesso Reflects patient behaviors

LimitsoSelection of indicator drugs : specificity oUse of medications vary by demographic factorsoPopulation source not precisely known : incidence ?

EHESP

13

o Relevant tool to determine dynamics and detect outbreaks o Reporting lag of one day

o rapid assessment of Public Health situationo prospective analyses

o Automatically collection of data

Advantages

Conclusion

Useful and valid tool for real-time monitoring of GI

Earlier indicator of gastroenteritis outbreak

Other infectious diseases

EHESP

14

Thank you

QUESTIONS ?

EHESP

15

EHESP

16

Epidemics detection

o Detection Method (Serfling method) :

• Periodic regression models • Key parameters :

• highest pruning percentile (varying from 15% to 40%)

• prediction interval (varying from 90%,95%,99%)• Number of consecutive weeks to detect an

epidemic

17

ANNEXES

EHESP

18

The selected detection model for non-prescribed drugs allows the detection of seasonal outbreaks 2.25 weeks earlier

Detection performance of the selected model (IC 95%, cut-off 30%)Sensitivity : 100%False alert rate : 0%Mean timeliness: -2.25 weeks (min -3; median -2.5, max -1)

EHESP

19

The selected detection model for prescribed drugs allows the detection of seasonal outbreaks 0.6 weeks earlier

Detection performance of the selected model (IC 99%, cut-off 30%)Sensitivity : 100%False alert rate : 0%Mean timeliness: -0.6 weeks (min -2; median 0, max +1)