Sequential methodology in case series analysis: The UK experience

1
Vaccine safety evaluation post marketing
surveillance conference Bethesda, Maryland April
10-11, 2007

• Sequential methodology in case series analysis
  The UK experience
• Mounia Hocine1, P. Musonda1, N. Andrews2, C.P.
  Farrington1
• 1 The Open University, Milton Keynes, UK
• 2 Health Protection Agency, London, UK
• Biostatistician, PhD (m.hocine_at_open.ac.uk)

2
The self-controlled case series method (SCCS)

• An epidemiological study design to investigate
  the association between apoint exposure and an
  acute event.

• The method uses only cases, and estimates the
  relative incidence of the event in a
  pre-defined post-vaccination risk period.

• Widely used in studies of vaccine safety.

• So far the method has been used retrospectively,
  but it may also be adapted for use in
  prospective surveillance.

3
The SCCS method
(Farrington, Biometrics 1995)
For each case i
Calendar period
ai
bi
nijk Poisson(eijk lijk)
? Multinomial likelihood(?i factor out)
4
The SCCS method Advantages

• Uses data from only cases,

• Adjusts for all fixed confounders over time and
  for exponential trends in incidence in
  calendar time,

• Powerful, quick and easy to apply, when
  applicable

• Applicable to databases from hospitals, General
  Practice Research Database (GPRD),

5
The SCCS method Surveillance scenario

• Focus on specified adverse events after the
  introduction of a new vaccine,

• The analysis is undertaken prospectively at
  regular surveillance intervals (3, 6 or 12
  months),

• At the end of each interval the vaccination
  records of the cases notified in that
  interval are ascertained (such data might be
  obtained from hospital records, the GPRD,)

6
Sequential Probability Ratio Test

• H0 RI 1 vs H1 RI 1.5, 2, 5, 10,

• Apply the SCCS at the end of each surveillance
  interval i ? L0(Xi), L1(Xi),

• We define two boundaries U and L as functions of
  a and b
• Upper boundary U log (1-b)/a,
• Lower boundary L log b/(1-a).

7
SPRT Decision rule
SPRT
U
Continue
0 1 2 3
4 5 6
7
Time
L
STOP
H0 accepted
8
Surveillance scenario simulation

• Properties of the case series SPRT were studied
  by simulation,

• We mimicked a vaccination programme targeted at
  1-2 year olds, with most vaccinations
  occurring at 12-18 months of age,

• We investigated the effects of age, risk period,
  surveillance interval, relative incidence
  (under H1).

9
(No Transcript)
10
(No Transcript)
11
Illustration Bells palsy and parenteral
inactivated influenza vaccine using the UK GPRD

• Bells palsy is an acute facial paralysis
  affecting the 7th facial nerve.

• Most cases report to their GP.

• Association reported with an intranasal flu
  vaccine OR 84.0 (Switzerland), but not with
  parenteral vaccine.

• Further studies undertaken using data from the
  GPRD (no association)

12
Illustration Bells palsy and flu vaccineusing
case series SPRT

• Retrospective application of case series SPRT
  from July 1992

• Temporal adjustment by calendar month

• Risk period 1-7 or 1-60 days after any dose of
  influenza vaccine

• H1 RI 1.5 and 5

• Surveillance interval length 6 months

• Degree of urgency alarm a b 0.01 ? L -4.6
  (lower boundary)

13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
Conclusion

• The self-controlled case series method
  provides a robust and readily implemented method
  for routine surveillance.
• Website
• http\\statistics.open.ac.uk\sccs
• (Bibliography - Programs in STATA, SAS, GENSTAT
  and GLIM)