Futures generation: Modelling healthcare need, activity and outcomes

1
Futures generationModelling healthcare need,
activity and outcomes

• Georgios Lyratzopoulos,
• MFPH, MRCP, MPH, DTMH

2
Contents

• Basic concepts
• Modelling outcomes case studies
• HF specialist clinic
• GP contract CHD targets
• Modelling activity outcomes
• IUI vs. IVF comparison
• Modelling need (health determinants)
• Metastatic liver disease

3
What is modelling

• A simulation reality, however, reality is
  idiosyncratic
• Explanatory (historical)
• HDA smoking prevalence / PAR in small areas
• McPherson et al. BHF report,
• Capewell et al.
• Predictive (futures generation)
• Roderick and Davies Renal (NSF) model
• Capacity models (several unpublished examples)

4
IMPACT CHD Model EW 1981-2000 Capewell et al.,
several papers, including Circulation, Heart, Eur
J Cardiol
Risk Factors worse 15 Obesity (increase)
2 Diabetes (increase) 5 Physical
activity (less) 8 Risk Factors better
-71 Smoking -31 Cholesterol
-12 Population BP fall -16 Deprivation
-4 Other factors
-8   Treatments -44 AMI treatments
-6 Secondary prevention - 10 Heart
failure -11 AnginaCABG PTCA
-4 Angina Aspirin etc
-7 Hypertension treat. -8
80,500 fewer deaths
2000
1981
5
Relative importance of primary care avoidable
mortality in lt75 year olds
Tobias M and Jackson G. ANZJPH 2001
6
Why model Literature not enough, primary
research not feasible (attributed to I Harvey)
Read (literature review and critical appraisal)
Model (simulate care pathways, patient flows,
outcome and costs)
Study (audit, health services research.)
7
Modelling healthcare applications
Need
Activity
Outcomes
(Very similar to modelling the distribution of a
health determinant)
Costs and benefits
8
Case study 1

• Modelling expected outcomes of a new service
  (Heart Failure Specialist Clinic)

9
Effective non-invasive treatments for HF
 
10
Parameters of the particular impact assessment
exercise

• Specialist HF services sum of three constituent
  interventions (i.e. spironolactone, b-blockers,
  N-LEI)
• Only modelling impact of services covering HF
  patients with previous hospitalisations
• Only modelling impact on mortality and readmission

11
Modelling of potential impact of a single
intervention

• NPE(a, t) n Pe-u(a) pt (not on a) RRR(a)
• n number of patients with condition (HES)
• Pe-u proportion of eligible but untreated
  patients (literature and audit)
• pt probability of death or mean number of
  re-admissions per patient during t (HES, HES
  ONS mortality file)
• RRR relative risk reduction associated with
  treatment (meta-analyses or large RCT)

12
Proportion of patients eligible but untreated
P e-u 1 ( P treated P ci / intolerant )
13
Pe u
14
Risk of death and mean number of readmissions

• Annual risk of death 32
• Mean annual number of readmissions / patient
  0.77
• Assumption The observed values for all patients
  are also applicable to patients not on b-blockers
  and spironolactone

15
Spironolactone / deaths potentially prevented
during one year

• NPE n Pe-u(a) p(not on a) RRR(a)
• NPE (spir) 286 0.55 0.32 0.31
• NPE (spir) 14
• (16 of all 90 expected deaths)

16
Calculating the impact of combination therapies
(Mant Hicks formula)

• (RRab) 1 (RRRa) (RRRb) .
• Assumption 1 Eligibility for one intervention is
  independent of eligibility for any other
• Assumption 2The effect of nurse-led education is
  independent of improved compliance with
  b-blockers and spironolactone (Model 3)

17
Modelling expected impact from investment in
specialist services (heart failure)
18
(No Transcript)
19
Case study 2

• Modelling expected outcomes of a new policy (2003
  GP contract)

20
QOF CVD quality targets(NB no organisational,
no tobacco)
21
Requirements
Effectiveness of interventions Recently
published meta-analyses or large RCTs
(RRR) Current risk factor burden / treatment
uptake 1998 HSE for cholesterol and BP levels
Northumberland primary care data collection
project for treatment uptake and other
population-based sources
22
Requirements (cont.)
Prevalence of CHD, Stroke, Hypertension and
Diabetes HSE 1998 Baseline risk Using
Framingham equations on 1998 Health Survey for
England participants
23
Calculation of Number of Events Prevented for a
therapy
NEP n ? pr ? (1-ci) ? pe ?
P ? RRR n no. of people in population
of interest pr prevalence of the disease in
the population ci proportion of patients with
a contraindication pe incremental increase in
the use of the treatment P probability of the
outcome of interest (baseline risk) RRR
relative risk reduction associated with the
treatment
Similar approaches Heller RF et al. BMC Public
Health. 200337. Lyratzopoulos G et al. BMC
Health Services Research, 2004410
24
NEP n ? pr ? (1-ci) ? pe ? P ?
RRRExample of optimising aspirin uptake to 90
among men with CHD, 45-64 y

• n
• 1,154
• n pr
• 1,154 X 8 96
• n pr 1-ci
• 1,154 X 8 X 90 87
• n pr 1-ci pe
• 1,154 X 8 X 90 X 9 8
• n pr 1-ci pe P
• 1,154 X 8 X 90 X 9 X 13 1
• n pr 1-ci pe P RRR
• 1,154 X 8 X 90 X 9 X 13 X 0.25
  0.25 (NEP)

25
NEP n ? pr ? (1-ci) ? pe ? P ?
RRRExample of optimising aspirin uptake to 90
among CHD patients men 45-64

• n
• 1,154
• n pr
• 1,154 X 8 96
• n pr 1-ci
• 1,154 X 8 X 90 87
• n pr 1-ci pe
• 1,154 X 8 X 90 X 9 8
• n pr 1-ci pe P
• 1,154 X 8 X 90 X 9 X 13 1
• n pr 1-ci pe P RRR
• 1,154 X 8 X 90 X 9 X 13 X 0.25
  0.25 (NEP)

26
NEP n ? pr ? (1-ci) ? pe ? P ?
RRRExample of optimising aspirin uptake to 90
among CHD patients men 45-64

• n
• 1,154
• n pr
• 1,154 X 8 96
• n pr 1-ci
• 1,154 X 8 X 90 87
• n pr 1-ci pe
• 1,154 X 8 X 90 X 9 8
• n pr 1-ci pe P
• 1,154 X 8 X 90 X 9 X 13 1
• n pr 1-ci pe P RRR
• 1,154 X 8 X 90 X 9 X 13 X 0.25
  0.25 (NEP)

27
NEP n ? pr ? (1-ci) ? pe ? P ?
RRRExample of optimising aspirin uptake to 90
among CHD patients men 45-64

• n
• 1,154
• n pr
• 1,154 X 8 96
• n pr 1-ci
• 1,154 X 8 X 90 87
• n pr 1-ci pe
• 1,154 X 8 X 90 X 9 8
• n pr 1-ci pe P
• 1,154 X 8 X 90 X 9 X 13 1
• n pr 1-ci pe P RRR
• 1,154 X 8 X 90 X 9 X 13 X 0.25
  0.25 (NEP)

28
NEP n ? pr ? (1-ci) ? pe ? P ?
RRRExample of optimising aspirin uptake to 90
among CHD patients men 45-64

• n
• 1,154
• n pr
• 1,154 X 8 96
• n pr 1-ci
• 1,154 X 8 X 90 87
• n pr 1-ci pe
• 1,154 X 8 X 90 X 9 8
• n pr 1-ci pe P
• 1,154 X 8 X 90 X 9 X 13 1
• n pr 1-ci pe P RRR
• 1,154 X 8 X 90 X 9 X 13 X 0.25
  0.25 (NEP)

29
NEP n ? pr ? (1-ci) ? pe ? P ?
RRRExample of optimising aspirin uptake to 90
among CHD patients men 45-64

• n
• 1,154
• n pr
• 1,154 X 8 96
• n pr 1-ci
• 1,154 X 8 X 90 87
• n pr 1-ci pe
• 1,154 X 8 X 90 X 9 8
• n pr 1-ci pe P
• 1,154 X 8 X 90 X 9 X 13 1
• n pr 1-ci pe P RRR
• 1,154 X 8 X 90 X 9 X 13 X 0.25
  0.25 (NEP)

30
Risk factor targets Simulation of expected risk
reduction until each individual reached target
Blood pressure target of lt 150/90. Women 65
to 84 years with hypertension but no history of
stroke, diabetes or CHD
31
Number of events prevented by target (10,000
population, 5-year period)
32
Number of CVD events prevented in a "typical"
population of 10,000 over 5-years from meeting
selected (grouped) GP contract targets
33
(No Transcript)
34
Case Study 3

• Modelling activity (outcomes) cost
• (IVF, vs. IUI IVF)

35
Flow chart of couples offered IUI (mutually
eligible for IUI and IVF)
IUI costs
Total cost
IVF costs
36
Cost and cost-effectiveness (/ Live Birth) of
different uptake of IUI and S-IUI among a
hypothetical cohort of 100 couples who are
mutually eligible for both IUI modalities and
IVF. Assumes constant LBR of 7 and 3.5 for
S-IUI and IUI
37
Case study 4

• Modelling health need for an intervention
  (surgery for metastatic liver disease due to
  colorectal primaries)

38
Number of primary colon and rectal cancer cases,
in East of England, by year 1991-2001.
Three-year averages of cases 1999-2001 were used
in this paper to estimate the number of cases in
2000, and multiplied by 10 to derive number of
operable cases
39
Metastastic liver cancer, East of England. Year
2000, 2005 and 2010 estimated number of
operable cases
40
Annual Mean Number of FCEs, Liver Excision /
Extirpation, 2000-2003
41
Role of sensitivity analysis

• CI in each factor
• Some assumptions based on evidence (/-
  generalisable), some on consensus or intuition
• Sensitivity analysis the answer

42
Sensitivity analysis

• Probabilistic, e.g. Bootstrap methods (Buchan
  http//simph.man.ac.uk/pinert/ )
• Scenario-based (quick and perhaps not as bad, see
  Capewell).

43
(No Transcript)
44
Modelling healthcare applications
Need
Activity
Outcomes
Liver Excisions
Heart Failure IUI vs. IVF
GP contract Heart Failure IUI vs. IVF
Costs and benefits
45
Some truisms for the end

• The need to model is at least as great and
  frequent as the need for literature-based
  evidence and primary research
• There are plenty of modelling methodologies out
  there see one / do one
• Can be time consuming / serious business
• Understanding of real care pathways critical
  key informers or direct observation / experience
  useful
• Policy makers appreciate it
• Good for primary research publication