Real-Life Research Seminar 25June12

Seminar: “Real-life” Research for

“Real-life” Research

Monday 25th June

RIRL

09:30 Welcome and introduction to real-life research Prof David Price

10:00 Practical examples of pragmatic and observational studies including

new models of pragmatic trials Prof David Price

10:45 Utilising databases for observational research Julie von Ziegenweidt

(Database manager)

11:30 Tea and coffee break

11:45 Managing confounding in observational databases. Adjustment,

matching and other techniques

Annie Burden

(Senior Statistician)

12:30 Lunch

13:15 Cardiovascular disease risk and pharmacological smoking cessation

interventions: a retrospective, real-life evaluation

Dr Erika Sims

(Senior Researcher)

14:00 Creating your own database – examples from practice iHARP – a 6

country cross-sectional database

Stan Musgrave

(Senior Research Fellow UEA)

14:45 Tea and coffee break

15:00 Practical group work – design your own project

16:00 Publishing real-life research Prof David Price

Program

Introduction to Real Life Research

Professor David Price

• What is real-life research?

• What are the advantages of real-life research

• How do we conduct real-life studies at RIRL

Introduction to Real-life research

Types of Trial

• Randomised Control Trial

o Patients with no confounding factors allocated at random to

receive different clinical interventions

o Measure efficacy

• Pragmatic Trial

o Randomised trial to compare two or more medical interventions

directly relevant to clinical care or health care delivery and

assess those interventions‟ effectiveness in real-world practice.

• Observational Study

o Case-control and cohort studies

o Retrospective analysis of outcomes to clinical interventions

carried out in normal patient care

What is effectiveness?

Population

Effectiveness

Efficacy x

“real-life” population of patients x

“real-life” management x

Understanding Clinical Effectiveness

What evidence do we really need?

Evidence

Theoretical

Theoretical model provide

rationale

Classical double-blind

double-dummy RCTs

Gold standard, large range of

outcomes.

But not “real-life” patients,

compliance and represent <10%

of patients

Pragmatic trials

More real-life Broader inclusion

criteria Allow normal factors to

occur Usually randomised

Simple outcomes But still consent &

rigorous

Observational Data

Real-life patients Not randomised

Routine data Normal decisions Difficult to ensure

group comparability

Matching of case controls,

adjustment

• What is real-life research?




Randomised Controlled Trials in Asthma as an Example

Travers et al. Thorax 2007

INCLUSION CRITERIA ECOLOGY OF CARE

Lung function 50 to 80% predicted Intense Patient Monitoring

Bronchodilator reversibility Regular inhaler review

Absence of co-morbidity Weekly assessments

Non smokers or ex-smokers <10 pack years

Daily diary of treatment

Symptomatic and regular use of medication

Good inhaler technique

High adherence to medication

Norwegian study of asthma patients to

identify who would be eligible for standard

clinical trials N

um

ber

of

Pati

en

ts All patients

Clinical asthma

FEV 50-85%

Reversibility 12%

No co-morbidity

Pack-year < 10

Reg use of ICS

Symptomatic asthma1.2%

Herland K, Akselsen JP, Skjonsberg OH, Bjermer L. How representative are

clinical study patients with asthma or COPD for a larger "real life" population of

patients with obstructive lung disease? Respir Med. 2005;99:11-9

External validity of RCTs

0

5

10

15

20

25

30

35

40

% indiv

iduals

pote

ntially

meeting

elig

ibili

ty c

rite

ria

References cited in GINA with level A or B evidence grade

Median 4%

Travers et al. Thorax 2007

n=179

Does it matter (even result in bias) if we exclude patients with:

• Poor inhaler technique

• Active rhinitis

• Smokers

• Lower adherence

• Lesser reversibility than 20%

AND

Design the study not like real-life?

Study only lasts 3 months?

Recommendations in guidelines often

muddled by “irrelevant” RCTs

Most inhaler device studies only include those with good inhaler technique!

Studies of real-life inhaler technique

References Inhaler device

1)Erickson et al. MDI

2)Larsen et al. MDI

3)Schmid et al. DPI (Turbuhaler®)

4)Shrestha et al. MDI

5)Thompson et al. MDI

6)van Beerendonk

et al.

MDI (19% ) or DPI

(81%)

82

58 64

79 76

89

0

10

20

30

40

50

60

70

80

90

100

1 2 3 4 5 6

Pe

rce

nta

ge o

f p

atie

nts

Reference number

Percentage of patients with inadequate inhaler technique

Reviewed in Duerden M, Price D. Dis Manage Health Outcomes 2001; 9 (2)

Subanalysis of Asthma Patients with Concomitant

Allergic Rhinitis in COMPACT

*Montelukast 10 mg once daily + budesonide 400 µg twice daily; **Budesonide 800 µg twice daily Price DB et al. Allergy 2006

Ch

an

ge

fr

om

bas

elin

e

(L/m

in,

LS

mea

n

SE

M)

p=0.36

Weeks

All patients

Montelukast Provided Greater Improvements in Asthma Patients with

Concomitant Allergic Rhinitis

50

40

30

20

10

0

Montelukast (n=433)*

Budesonide (n=425)**

0 4 8 12 Weeks

Montelukast (n=216)*

Budesonide (n=184)**

0 4 8 12

Patients with rhinitis 50

40

30

20

10

0

p<0.03

Smoking in asthma and effect of inhaled

steroids

Chalmers GW et al. Thorax 2002;57:226–230

Kerstjens HA, et al. Eur Respir J 1993;6(6):868-76.

NS + ICS

NS + PL

S + PL

S + ICS

High dose ICS (FP 500) vs LTRA in light

smokers

• 20-30% of asthma

patients smoke

• Studies of low

dose ICS have

shown little

efficacy

• This study

compared FP 250

bid, montelukast

10mg and placebo

over 6 months

Percentage of Days without Nocturnal Awakenings

Month

0 1 2 3 4 5 6 7

% o

f d

ays w

ith

ou

t n

octu

rna

l a

wa

ke

nin

gs

50

55

60

65

70

75

80

Montelukast

Fluticasone

Placebo

Price D et al ERS 2011

Monteleukast

ICS

Placebo

RCT References

1) Pawels R et al. N Engl J Med 1997

2) Kips J et al. Am J Respir Crit Care 2000

3) Bateman E. Am J Respir Crit Care 2004

4) Papi A et al. Eur Respir J 2007

5) Busse W et al. J Allergy Clin Immunol 2008

Real-world References

1) Partridge Pulm Med 2006

2) De Marco et al. Int Arch Allergy Immunol 2005

3 and 4) Janson et al. Eur Respir J 2001 3=Italy 4=UK

5 and 6) Breekveldt-Postma et al. Pharmaco-epidemiol Drug Saf 2008 5=fixed combination 6=ICS

7) Stallberg et al. Resp Med 2003

8) Adams et al. J Allergy Clin Immunol 2002

9) Corrigan Prim Care Resp J 2011

Real-life adherence in observational studies vs. randomised trials

0

20

40

60

80

100

1 2 3 4 5

Pe

rce

nta

ge o

f P

atie

nts

Randomised Trials 75-125 75-125

89 >95

>80

45

34

17

49

14.1 8.3

34

21

40

0

20

40

60

80

100

1 2 3 4 5 6 7 8 9

Real-world Trials

Long-term adherence in a clinical trial with inhaled steroids in children

Jonasson G, Carlsen KH, Mowinckel P. Arch Dis Child. 2000;83:330-3

Budesonide bid Placebo bid

• What is real-life research




RIRL Study Design

Index prescription date

(IPD) either:

•Initiation of treatment

•Step-up treatment

Baseline period

• matching cohorts

• confounding factor definition Outcome period

• outcome comparison

• adjusted for baseline confounders

0 -12m +12m Treatment option 2

Treatment option 1

• General Practice Research Database (GPRD)

o De-identified longitudinal data

o 500 primary care practices in the UK

o 13 million total records, of which 3.6 million are active

o Broadly representative of UK patient population

• Optimum Patient Care Research Database (OPC-RD)

o Anonymous data extracted from practices during records-

or nurse based clinical reviews

o Electronic records from over 500 000 patients with

respiratory disease

Clinical Datasets

GPRD (now the CPRD) website resources

• UK GP Practice Recruitment

• Identification of eligible patients according to protocol requirements

• Collection of routine data required for CRF direct from electronic patient record

• Pre-filling of CRF using routine data

• Nurse delivered consultations

Optimum Patient Care Research Service

• Benefits to GP

o Minimises impact of participating in study on workload

o Increases likelihood to participate

• Benefits to Client

o Reduces time spent in practice identifying eligible

patients

o Reduces time spent completing CRFs

o Reduces transposition errors incurred during

completion of CRF

Optimum Patient Care Research Service

Examples of Study Design 1

Patients enrolled & consented

into Study by OPC Research

Nurse

GPs Review Eligible Patient List

Invitations to participate sent to

eligible patients by remote

mailing

Extraction of Routine

Data from practice

management system

Identification of patients by

Read Code

Identify eligible patients

Report to GP

Anonymise

Anonymisation /

de-anonymisation

programme

De-anonymise

Export Routine data

into study template

Responses returned to OPC‟s

offices. Appointments made by OPC

Research Nurse to attend study day

at GP practice Routine Data

used to pre-fill

OPC Screening

Check List to aid

completion

alongside CRF OPC Screening Checklist (SC)

completed (including confirmation of

pre-filled data). Data from SC

transcribed into CRF.

• International multi-centre observational study

• Eligible patients identified using data from electronic patient record

• Paper CRF required to be completed at all sites for all data

• Screening Checklist enables completion of 58 of 70 (>80%) fields on CRF with data from electronic patient record

• OPC staff undertake study visits


Data prefilled from electronic patient record


International multi-centre observational study Eligible patients identified using data from electronic patient record Current therapy data directly exported as ASCII file to CRO for analysis Paper CRF completed for required to be completed at all sites Screening Checklist enables completion of 58 of 70 (>80%) fields on CRF with data from electronic patient record

Extraction of Routine

Data from practice

management system

Identification of patients by

Read Code

Identify eligible patients

Report to GP

GPs undertake patient visits

Anonymise

Anonymisation /

de-anonymisation

programme

De-anonymise

Export Routine data

into study template

OPC Electronic Screening Checklist

(ESC) prefilled; to be confirmed by patient during visit

GPs Review Eligible Patient List

Invitations to participate sent to

eligible patients by remote mailing

Routine Data

used to pre-fill

OPC Electronic

Screening

Check List

(analogous to

eCRF)

Export data for

recruited patients

into ASCII file for

direct entry into

study database


Data extracted from routine

patient record into OPC-RD

Prefill Electronic

SCL & Study

Database

Additional data directly into

electronic SCL & Study Database

Research staff do not need access to patient records minimising transposition

errors

Rapid data collection

• 12 European Countries, 163 sites in total

• 8150 patients in total (approximately 50 per site)

The role of OPC

• 10 sites needed in the UK

• Sponsor identified 4 sites

• OPC identified 6 sites

A prospective observational study in

asthma and OPC

• OPC has established a good relationship with a

number of practices across the UK

• Many practices are already research pro-active

and recognise the value of “Real-Life” Research

• Practices often involved in multiple studies with

OPC

• Practice involvement is rewarded by OPC‟s Clinical

Review Service and/ or remuneration

The Advantages of OPC

Practical Examples of pragmatic

and observational studies


• Recent papers from RIRL

• The ELEVATE trial

• Current Research

Practical Examples of Studies

Baseline: 1 year

Outcome: 1 year

Index prescription

date (IPD)

Matched 1:1 on baseline

demographics and

disease severity

Fluticasone

5 to 60 years Receiving a first prescription of ICS

J ALLERGY CLIN IMMUNOL SEPTEMBER 2010

0

5

10

15

20

25

30

35

40

45

50

Pe

rce

nta

ge o

f P

atie

nts

Dose of ICS at IPD

Fluticasone

HFA-BDP(Qvar)

HFA-BDP (Qvar)

Odds Ratios/Rate Ratios

0.7 0.8 1 1.2 1.5 2

Asthma control + SABA adjusted

Asthma control + SABA

Asthma control adjusted

Asthma control

Initiation Population: Odds Ratios and

Rate Ratios

*Adjusted for year of index date, acetaminophen, asthma consultations, rhinitis diagnosis, recorded

asthma diagnosis, and cardiac disease diagnosis

SABA: Short acting β-agonist

Price et al. J Allergy Clin Immunol 2010;126:511-8

FP set to 1.00

Greater asthma control for all definitions of

“control” with HFA-BDP (Qvar) compared to FP

95% CI

n = 1319 for both cohorts

Baseline: 1 year

Outcome: 1 year

Index prescription

date (IPD)

Matched 1:1

HFA-BDP (Qvar)

CFC-BDP Patients 5-60 years

Study Period: January 1997 to June 2007

0.7 0.8 1 1.2 1.3

Change in therapy

Exacerbations

Asthma control no therapy change

Asthma control + SABA

Proxy asthma control

Adjusted odds ratios and rate ratios for EF

HFA-BDP compared to CFC-BDP (95% CI)

CFC–BDP set to 1.00

INITIATION POPULATION

0.3 0.5 1 2 3

CFC–BDP set to 1.00

STEP-UP POPULATION

EF HFA-BDP n=2882

CFC-BDP n=8646

EF HFA-BDP n=258

CFC-BDP n=516

Fine-particle BDP gives greater asthma control than the larger particle

CFC-BDP – greater penetration into airways

Baseline: 1 year

Outcome: 1 year

Index prescription

date (IPD) Pressurized metered-dose inhaler (pMDI)

For patient characterisation

For outcome evaluation

Matched

Breath-actuated MDI (BAI)

Dry powder inhaler (DPI)

David Price John Haughney Erika Sims Muzammil Ali Julie von Ziegenweidt Elizabeth V Hillyer Amanda J Lee Alison Chisholm Neil Barnes

BAI and DPI (pMDI set to 1.00)

INITIATION COHORT

STEP-UP COHORT

pMDI (n = 39,746)

BAI (n = 9809)

DPI (n = 6792)

pMDI (n = 6245)

BAI (n = 1388)

DPI (n = 1536)

• Recruiting patients for study

o Asking GPs to recruit patients required

• Pre-defined database will make it easier in future to

carry out large-scale database studies

• Working with practices allows us to identify patients

for research in current studies

Challenges of the ELEVATE Trial

0 2 10 26 52 78 104

Week Week:

Tailored treatment as indicated by

guidelines

LTRA Ideally no ICS use

ICS Ideally no LTRA use

Baseline

V1 V2 V3 V4 V5 V6 V7

SABA

Randomisation

Study included to show that effectiveness results maybe difference to efficacy results; this study does not advocate the use of LTRA outside of the licence guidelines

CONSORT Diagram

* = Mean (SD), NR = not reported, NA = not applicable, %PPEF= Percent predicted PEF, %PFEV1= Percent predicted FEV1

1- Bateman et al. Can Guideline-defined Asthma Control Be Achieved?: The Gaining Optimal Asthma ControL Study Am. J. Respir. Crit. Care Med. 2004; 170: 836-844. 170. p.836, (2004)

ELEVATE

Step 2; N=306

GOAL

Strata 1; N=1098

Sex (% Female) 51% 57%

Age * 45.8 (16.4) 36.3 (15.6)

Quality of Life (Juniper AQLQ 1, worst, to 7)

4. 74 (1.04) 4.4 (1.00)

Lung Function * 86

%PPEF 77

%PFEV1

Percent reversibility * 8.9% (9.86) 22% (12.2)

Smokers – current 21.9% 9.5%

Drop out rate 4.0% 15.4%

Demographics and drop out rates Comparison to other studies (ELEVATE step 2, GOAL1)

First-Line Controller Therapy Trial Months

First-Line Controller Therapy

Trial

LTRA (n=148) ICS(n=158) Rate Ratio 95% CI LTRA vs. ICS

Mean no. of exacerbations 0.44±0.94 0.35±0.95 1.27 (0.83-1.92)

Secondary Outcome Measures

0

10

20

30

40

50

60

70

80

90

0 1 >1

Pe

rce

nta

ge o

f P

atie

nts

Number of Exacerbations

LTRA(n=148)

ICS(n=158)

LTRA ICS

Rate

%

0

20

40

60

80

100 p=0.11

n=108 n=101

65

92

15

41

62

21

Adherence to Therapy

Sub-group analysis - smokers

-2 0 8 26 52 78 104

Week Week:

Tailored treatment as indicated

by guidelines

LTRA + ICS No LABA use

LABA + ICS No LTRA use

Baseline

V1 V2 V3 V4 V5 V6 V7

ICS +SABA

Randomisation

PE

F (

% o

f p

red

icte

d)

Step-up Therapy Trial

Conclusions

• Whilst efficacy based studies suggest ICS are more

efficacious that LTRAs

• Real-life randomised trial i.e. real patients and real

practice – equivalent clinical outcomes

• Neither option universally effective

• A real-life trial assessing the safety and effectiveness of Relovair (fluticasone/vilanterol) in COPD and Asthma compared with usual care

• 4000 COPD patients and 5000 asthma patients

• 1 year study duration

Endpoints

• Collected partly from integrated primary and secondary care datasets

• Patient symptoms over time

• Exacerbation of symptoms

• Contact with healthcare professionals for respiratory reasons

• Other medication required to control symptoms

• Exhaled nitric oxide can be used as a marker for

inflammation

• Devices that measure this inflammation can be used

to help with asthma monitoring

• Using proxy measures of outcome control to explore

whether eNO monitoring can help to improve

asthma management

Using eNO to monitor patient asthma

treatment

The eNO study: Full study design

Patients reviewed with FeNO monitoring


(IPD): initiation of FeNO

monitoring

Baseline period: 1 year Outcome period:

1) 6 months for monitoring to take effect

2) 12 months to measure outcomes

Patients reviewed with no FeNO monitoring

Research Question: Does using eNO to monitor patient asthma improve their asthma outcmes.

Preliminary results: Changes

before and after eNO monitoring

Patients reviewed with FeNO monitoring


(IPD): initiation of FeNO

monitoring

Baseline period: 1 year Outcome period: 6 months

Patients must have •> 2 years history of asthma (diagnostic code and/or prescription for asthma therapy at least 2 years prior to IPD) •Evidence of current asthma treatment (≥2 asthma prescriptions during baseline + outcome year) •Have at least one year of up-to-standard (UTS) baseline data and at least 6 months of UTS outcome data (following the IPD).

Reasons for using eNO monitoring

Poor ICS adherence (MPR < 80%)

58.0%

Diagnosed not asthma no ICS in outcome period

21.8%

Evaluation of high risk (>1 exacerbation in baseline)

18.3%

General Monitoring

36.4%

Data Source: Optimum Patient Care Research

Database

• Anonymous data extracted from practices for

chronic respiratory service review.

• Two types of data collected

o Routine Clinical Data

o Patient Recorded Data

• OPCRD has been approved by Trent Multi Centre

Research Ethics Committee for clinical research use

Preliminary Results: Patient Demographics

eNO Treatment Group

BMI (kg/m2) 25.7 (23, 30.4)

Female n(%) 196 (52.8)

Smoking Status

Non 260 (70.1)

Current 19 (5.1)

Ex 86 (23.2)

Missing data

6 (1.6)

Frequency distribution for age Mean: 42.26 S.D: 20.77

Preliminary results: Co-morbidities and

Therapies

Co-morbidities* in eNO monitoring group n(%)

Rhinitis 174 (46.9)

GERD 20 (5.4)

Cardiac Disease 6 (1.6)

Medication therapies+ in eNO monitoring group n(%)

Beta Blockers 4 (1.1)

NSAIDS 196 (52.8)

Paracetamol 57 (15.4)

*Defined as Read code for co-morbidity at any time

+Defined as prescriptions received during the 1 year prior to IPD or at IPD

Sub Analysis: Patients increasing their average daily

ICS dose from baseline to outcome by ≥50%

Patients prescribed in eNO Monitoring Group n(%) Mean (SD)

Before eNO monitoring

Following eNO monitoring

Based on 6-month outcome period (Counts scaled to 365 days)

Total Acute Oral steroids 0.26 (0.75) 0.32 (1.13)

LRTI Consultations resulting in script for antibiotics

0.08 (0.29) 0.05 (0.38)

Based on 1 year outcome period

Total Acute Oral steroids 0.21 (0.64) 0.23 (0.63)

LRTI Consultations resulting in script for antibiotics

0.12 (0.35) 0.05 (0.26)

Change in Asthma Severity by BTS Step

0

5

10

15

20

25

30

35

0 1 2 3 4

Pe

rce

nta

ge o

f P

atie

nts

BTS Stage

Baseline

Outcome

Preliminary Results: Change in Control Status

Baseline Patients n(%) Outcome Patients n(%)

Uncontrolled 83 (22.4) 47 (12.7)

Controlled 288 (77.6) 324 (87.3)

Total 371 (100) 371 (100)

77.6 87.3

0

20

40

60

80

100

Baseline Outcome

Patients with fully controlled asthma before and after eNO monitoring

Utilising databases for observational

research

Julie von Ziegenweidt

Database Manager

RIRL

• Julie von Ziegenweidt o Senior Analyst/Data Manager

• Jeremy Brockman o Data Analyst

• Daniel West o Junior Analyst

RiRL Data Analysis Team

• Different types of studies

• Study Design

• Choosing the right data source

• Data Analysis

o Understanding the data

o Data Management

o Extracting the study data

o Quality Checks

o Matching and randomisation

o Study Archive

• Summary

Data Analysis

• Observational studies Comparing 2 study periods to determine if the intervention

make a difference?

• ie „to compare the effectiveness of QVAR v FP in Asthma‟

This may be Retrospective or Prospective

• Prevalence studies

o The total number of cases of a disease in a given population at a specific time.

– ie identify proportion of patients with asthma on long term oral steroids in the UK

Different types of studies


• Study Design


• Data Analysis


o Data Management


o Quality Checks


o Study Archive

• Summary

Data Analysis

Study Design

Index date either:

•Initiation of treatment

•Step-up treatment

Baseline period

• matching cohorts

• confounding factor definition Outcome period



0 -12m +12m Treatment option 2

Treatment option 1


• Study Design


• Data Analysis


o Data Management


o Quality Checks


o Study Archive

• Summary

Data Analysis

• Are they representative?

• Do they provide the right data to answer research question?

• Do they provide the data in the way the researcher can work with it?

Retrospective Data sources

• Bespoke database to answer a specific question

• Example: The iHARP database

Prospective data sources

Patient Questionnaire

Consultation with respiratory clinician

and patient

Data from all three sources present in

the database

Review of routine practise data


• Study Design


• Data Analysis


o Data Management


o Quality Checks


o Study Archive

• Summary

Data Analysis

Preparation for analysis

(Study protocol and design)

Study population

(Consort Diagram, Inclusion /Exclusion

criteria)

Matching

(Matching criteria)

Data Analysis

o Coding Practices

– How has the data been sourced? (clinical v

questionnaire)

– Multiple coding environments (primary care v

insurance)

o Coding dictionaries

– ICD9/10, UK Read codes v2/3, or ATC or NCD

o Understanding definition of each column of data

– Patient ID, EventDate etc

Understanding the data

o Code lists – Asthma – read code, read terms.

– ICS inhalers – code, name, strength, device (MDI,BAI,DPI), device types (easi-breathe), doses in pack, drug, form etc.

– Interpretation tables – 2 puffs twice a day = 4 doses a day

o Data validation: – Duplication of entries

• multiple entries on same day for same drug but different prescribing instructions.

o Interpretation of possible mis-diagnosis – sensitivity searches

o Cleaning of data: – Quality of data –

• registration date prior to year of birth

• recorded height for child is over 2m

Data Management

Consort Diagram

Patient on any ICS

Increase ICS dose

• The Consort Diagram is a simple flow diagram to show the progression of patient identification

• Confirmation that patient identification is in line with study design

• For Example:

Change to combination therapy via fixed

combination inhaler

Addition of a LABA to existing ICS

• One row per unique patient

• Perform descriptive stats to ensure correctness of data (ie, means,

medians, SD, perc)

• Data dictionary o descriptions,

o data types (i.e. string or integers),

o indicators of missing data,

o definition of proxies

Data Quality Verification

• In real world studies case-control matching can be necessary to ensure

similarity of patients included in an analysis to minimise outcome

confounding

Matching

Baseline characteristics descriptively compared to identify variances between

cohort severity

Cases are potentially matched to multiple

controls

Random case-control matches

MATCHING

RANDOMISATION

Example of Matching Criteria

MATCHING CRITERIA AND CATEGORIES

• Age: < 12 (exact match) ≥12 (+/- 5 yrs)

• Average Daily SABA Dose 0 – 200, 201 – 400, 401 +

• Sex: male / female

• Last ICS dose prior to IPD 1 –100, 101 – 200, 201 – 400, 401 – 600, 601 – 800, 801 +

• Baseline primary outcome status

• Number of asthma consultations

0,1,2+

Example: The ICS vs Combination Therapy

Study

Preparation for study

Study population

Matching

Outcomes: proxy for asthma control, exacerbations Covariates: Smoking etc. Inclusion/Exclusion criteria: Valid asthma, valid data

RESULT: Valid population of anonymous matched patients

• Study Archiving

o Raw Data

o Extracted Datasets

o Code Lists

o Drug Dictionaries

o Data Dictionary

o Program Scripts

Lastly...

• Exploration and familiarisation of raw patient data essential

• Data validation (check for inconsistencies etc.)

• Set up of Diagnostic and drug lookup tables

• Study design specifies all data required at the outset o ie ensuring outcomes are appropriate for study hypothesis.

• Consort diagram communicates accurately the methodology behind study design.

• Continuous Data validation checks!!

Summary

Managing Confounding in

Observational Databases

Annie Burden

Senior Statistician

RIRL

• Annie Burden

o Senior Statistician

• Vicky Thomas

o Lead Statistician

o SAS

• Francesca Barion

o Lead Health Economist

o STATA

• Muzammil Ali

o Statistician

o SPSS

RiRL Statistics Team Statistics & Health Economics

• RCTs

o Subjects are Randomised

o Variation in baseline characteristics should be random

across treatment groups

• Observational Studies

o We get what we get!

o Variation in the baseline characteristics between

treatment groups may be:

– Systematic

– Random

Introduction

“Any claim coming from an

observational study is most

likely to be wrong”.

S. Stanley Young & Alan Karr

Significance Magazine September 2011 Volume 8 Issue 3

The mis-conception

• Multiple Testing

o Keep asking questions of the data & something will

eventually come up positive;

o E.g. “Women eating breakfast cereals leads to more

boy babies”.

• Bias

o Hidden Confounders

• Multiple Models

o Limitless possible combinations of confounders.

The mis-conception (continued)

• Multiple Testing

o Limited number of Primary Outcomes;

o Outcomes set a priori;

o Interpretation is important...

– Clinician Input (Female diet cannot influence baby‟s gender)

• Bias

o Try to include all potential confounders.

• Multiple Models

o Very careful selection of:

– Potential confounders; and

– Covariates to include in the final model.

The truth

• “A confounding variable is an extraneous variable in a statistical model that correlates (positively or negatively) with both the dependent variable and the independent variable.”

• “... Confounding is a particular challenge.” Wikipedia

• “Confounding – where the estimated association is NOT the same as the true causal effect...”

Thomas Lumley 2005

• “Throwing things into disorder; mixing up; confusing.” The Concise Oxford Dictionary

Confounding - Definitions

Confounding

• Baseline / Characterisation Variables related to the

Outcome (Dependent) Variable

o Predictive Variables;

• Baseline / Characterisation Variables related to the

Treatment (Independent Variable)

o Baseline Differences;

• Check for relationships between the potential

confounders to avoid double-accounting

o Linear relationships through correlation coefficients;

o Non-linear relationships via modelling / plots.

Potential Confounders

Treatment Group P value

Control Treatment Total

Number of

Exacerbations (ATS

Definition)

N (% non-missing) 194 (100.0) 388 (100.0) 582 (100.0)

0.949 Mean (SD) 0.32 (0.60) 0.32 (0.60) 0.32 (0.60)

Median (IQR) 0 (0, 1) 0 (0, 1) 0 (0, 1)

Total acute Oral

steroids

N (% non-missing) 194 (100.0) 388 (100.0) 582 (100.0)

0.986 Mean (SD) 0.12 (0.35) 0.12 (0.33) 0.12 (0.33)

Median (IQR) 0 (0, 0) 0 (0, 0) 0 (0, 0)

LRTI Consultations

resulting in

prescription for

Antibiotics

N (% non-missing) 194 (100.0) 388 (100.0) 582 (100.0)

0.781 Mean (SD) 0.24 (0.54) 0.22 (0.54) 0.23 (0.54)

Median (IQR) 0 (0, 0) 0 (0, 0) 0 (0, 0)

Baseline Differences

Gender Height Weight BMI (categorised)

Gender

Height 0.54

Weight 0.34 0.55

BMI (categorised) 0.78

Linear Relationships

Spearman’s Correlation Coefficients

Non-linear Relationships

Example Output from SPSS Statistics19 www.spss.com/uk/software/statistics/

http://www.spss.com/uk/software/statistics/

2 main options: • Perform comparisons only between observations

that have the same value of the confounder:

o Stratify data by confounder(s)

o Adjust for a confounder(s) in regression (an

approximation to stratification that requires less data).

• Perform comparisons only between groups that

have the same distribution of the confounder:

o Match on the confounder(s)

Removing Confounding

Score Statistics For Type 3 GEE Analysis

Source DF Chi-Square Pr > ChiSq

Treatment 1 9.8 0.0017

Age 1 3.84 0.0501

Acute_Oral_Steroids 2 20.01 <.0001

Non_asthma_Consults 3 7.83 0.0496

ZYear_of_IPD 1 1.69 0.1941

Asthma_A&E 1 24.34 <.0001

Rhinitis_Dx 1 8.93 0.0028

Adjustment

Example Output from SAS v9.3 www.SAS.com/offices/europe/uk/technologies/analytics/statistics/stat/ondex.html

Aim for Parsimony!

http://www.sas.com/offices/europe/uk/technologies/analytics/statistics/stat/ondex.html

• To minimise baseline differences between treatment groups

• Match on:

o Demographic variables (Age, Gender);

o Study-appropriate measures of baseline disease severity, for example:

– Baseline Exacerbations

– Baseline Consultations

– Controller Medication

– Reliever Medication.

• Matching Ratio (e.g.1:1, 1:2, 1:3) to maximise power of statistical tests

Matching

• Minimal adjustment for residual confounders

o Parsimony is Good!

• “Conditional” Models

o Conditional Logistic Regression

o Conditional Poisson Regression

o Conditional Ordinal Logistic Regression

• Matching aims to minimise differences between treatment groups at baseline BUT....

• Need to consider whether matched cohorts are then representative of the full populations.

Matching

• Another option but not as readily understood by reviewers etc.

• Using covariates predictive of outcome, calculate Propensity Score (using Logistic Regression):

= P(Treatment | Covariates)

• Match on Propensity Scores

• Advantages

o Easily Includes Multiple Covariates

• Disadvantages

o Does not necessarily match clinically “similar” patients

Propensity Scoring

Asthma Control Treatment Group

Total Control Active

Controlled n (%) 267 (70.4) 540 (71.2) 807 (71.0)

Uncontrolled n (%) 112 (29.6) 218 (28.8) 330 (29.0)

Total n (%) 379 (100) 758 (100) 1137 (100)

Odds Ratio adjusted for baseline

confounders* (95% CI) 1.00

1.24 (0.92, 1.66)

Example Results ASTHMA CONTROL Defined as: Controlled: the absence of the following during the one-year outcome period: •Asthma-related:

oHospital attendance or admission oA&E attendance, OR oOut of hours consultations, OR oOut-patient department attendance

•GP consultations for lower respiratory tract infection •Prescriptions for acute courses of oral steroids. Uncontrolled: all others.

*Adjusted for: Number of exacerbations (clinical definition) (categorised), Number of non-asthma-related consultations (categorised) and GERD diagnosis &/or therapy (YES/NO).

Logistic Regression Model

Number of Exacerbations (ATS Definition)

Treatment Group Total

Control Active

None n (%) 306 (80.7) 613 (80.9) 919 (80.8)

1 n (%) 57 (15.0) 99 (13.1) 156 (13.7)

2+ n (%) 16 (4.2) 46 (6.1) 62 (5.5)

Total (n) 379 (100) 758 (100) 1137 (100)

Rate Ratio adjusted for baseline


1.04 (0.76, 1.44)

Example Results

EXACERBATIONS (ATS DEFINITION) Defined as the occurrence of: •Asthma-related:

oHospital attendance / admissions OR oA&E attendance

•Use of acute oral steroids.

*Adjusted for: GERD diagnosis &/or therapy (YES/NO), Number of Primary Care Consultations (categorised) and Number of prescriptions for SABA (categorised).

Poisson Regression Model

Adherence to ICS Therapy

Treatment Group

Total Control Active

< 50%n (%) 186 (49.1) 162 (21.4) 348 (30.6)

50-69% n (%) 53 (14.0) 190 (25.1) 243 (21.4)

70-99% n (%) 79 (20.8) 132 (17.4) 211 (18.6)

≥ 100% n (%) 61 (16.1) 274 (36.1) 335 (29.5)

Total n (%) 379 (100) 758 (100) 1137 (100)

Odds Ratio adjusted for baseline


2.79 (2.21, 3.52)

Example Results

ADHERENCE TO ICS THERAPY

*Adjusted for: Age, Asthma diagnosis (YES/NO), Number of Primary Care consultations (categorised), Number of prescriptions for SABA (categorised), spacer device use (YES/NO) and Year of IPD.

Ordinal Regression Model

• Treatment Effect varies with stratum of the 3rd variable

o E.g. Active Drug has different effectiveness relative to

Control depending on smoking status / gender / year

• Effect Modification & Confounding can exist separately

or together:

o Effect modification without confounding

– Adjust & Look at interactions

o Confounding without effect modification

– Adjust / match

o Both confounding and effect modification

– Adjust / match AND

– Look at interactions

Effect Modification

Effect Modification (cont.)

• Exploration of and familiarisation with Data very

important

o Data validation (check for inconsistencies etc.)

o Patient Characterisation & Baseline Differences

between Treatment Groups

o Variables Predictive of Outcome

o Relationships between Variables

• Interpretation of the results / Clinical Input

o Ensure results are sensible

o Ensure adjustments are sensible

Summary

Cardiovascular disease risk and

pharmalogical smoking cessation

interventions: a retrospective, real-

life evaluation Dr. Erika Sims

Senior Researcher

RiRL Research Team

-

-

-

Clinical Research Team RiRL Academic

Professor David Price RiRL Chief Investigator

Professor of Primary Care Respiratory Medicine University of Aberdeen

Dr Erika Sims RiRL Senior Researcher

Honorary Research Fellow University of East Anglia

Dr Stan Musgrave RiRL Research and Medical Data Associate

Senior Research Fellow University of East Anglia

Dr Yolande Cordeaux Medical Researcher Research Fellow; University of Cambridge

• Tobacco dependence is a chronic, relapsing

condition

Smoking cessation interventions


condition

• In EU in 2000

o 655,000 deaths attributed to tobacco use

o Societal & healthcare costs €97.7–130.3 billion



condition

• In EU in 2000



• Two approaches to smoking cessation:

o Smoking Cessation Advice

o Pharmacological support



condition

• In EU in 2000



• Two approaches to smoking cessation:

o Smoking Cessation Advice

o Pharmacological support

– Nicotine replacement therapy - since 1980‟s

– Bupropion (2000) & Varenicline (2006)


Nicotine Replacement Therapy

• Nicotine Replacement Therapy

o Substitutes for nicotine

o nasal sprays, inhalers, gum and tablets, transdermal

patches

Safety of NRT

• Gillies et al 2012; Intensive Care Medicine, in press

o Retrospective case review

o No evidence of „harm associated with NRT, with the ICU model actually trending towards benefit‟.

o n = 423

• Ruiz et al 2012; Nicotine & Tobacco Research, in press

o Prospective study of COPD patients

o „All types of treatments were safe.‟

o n = 472

• Zapawa et al 2011; Addictive Behaviours vol 36: 327

o Systematic Review

o „Persistent (i.e., long-term) use of NRT does not appear harmful‟

Preliminary analyses: NRT vs SC

Adjusted for history of CVD, age and sex:

o 1.44 (95% CI: 1.17–1.79) for CVD


• Adjusted for history of CVD, age and sex:

o 1.44 (95% CI: 1.17–1.79) for CVD

• No difference in cardiovascular risk profile :

o body mass index (BMI),

o hyperlipidaemia,

o systolic blood pressure,

o hypertension and

o diabetes.


• Adjusted for history of CVD, age and sex:

o 1.44 (95% CI: 1.17–1.79) for CVD

• No difference in cardiovascular risk profile : o body mass index (BMI),

o hyperlipidaemia,

o systolic blood pressure,

o hypertension and

o diabetes.

• But limited analysis.

o population not large enough to draw conclusion on mortality.

o Other confounders?

• Patients exposed to NRT and other smoking

cessation pharmacotherapies are at a higher risk of

CVD compared with patients undertaking quit

attempts unaided by pharmacotherapies.

Hypothesis

• Patients exposed to NRT and other smoking

cessation pharmacotherapies are at a higher risk of

CVD compared with patients undertaking quit

attempts unaided by pharmacotherapies.

• 4 way analysis:

–NRT Smoking Cessation

–Varenicline Advice

–Bupropion

Hypothesis

VS

Study Design


Initiation of NRT

Baseline period

•no CS pharmacological aids Outcome period



0 -12m +12m SC advice

NRT

• Retrospective, matched cohort study using GPRD

2006 - 2009

Study Cohorts

• Exposure Cohort – NRT o No recorded exposure to CS pharmacological aids in the prior year,

o First recorded smoking cessation intervention was NRT (using any of, or a combination of products) at the index date.

Study Cohorts



• Comparison Cohort o No recorded smoking cessation attempts using pharmacological aids in the prior year

o First recorded smoking cessation intervention was smoking cessation advice unaided by pharmacological therapies at the IPD and during the outcome periods.

Study Cohorts



• Comparison Cohort o No recorded smoking cessation attempts using pharmacological aids in the prior year

o First recorded smoking cessation intervention was smoking cessation advice unaided by pharmacological therapies at the IPD and during the outcome periods.

• All Patients o Aged: 18–75 years

o Current smoker throughout the prior year (any quantity of cigarettes).

o No past history of CVD

o Have at least one year of up-to-standard (UTS) baseline data as defined by GPRD (prior to the IPD) and at least 4 weeks‟ of UTS outcome data (following the IPD) or UTS data up to the time of death if death occurred within the outcome period.

Study Outcomes

• Cardiovascular event during 52-week outcome period:

o Coronary Heart Disease diagnosis

o Coronary Heart Disease related death

o Cerebrovascular disease diagnosis

o Cerebrovascular disease death and No of Days from IPD

o Number of GP consultations

o Hospital attendances for Coronary Heart Disease or

Cerebrovascular disease, (including admission, A&E attendance,

out-of-hours attendance, or Out-Patient Department (OPD)

attendance)

• Demographics

• Co-morbidities

• Therapies

• Clinical Outcomes

• Healthcare utilisation

• As for Baseline Variables

• Death

Baseline Variables

Statistical Analysis

• Baseline Variables

o Descriptive Analysis Means: t-test

Medians: Mann-Whitney U-Test

Proportion: Chi-squared test

• Matched Baseline & Outcome Variables

o Conditional Logistic Regression

• Change from Baseline Analyses

o Unadjusted: Conditional Logistic Regression

o Adjusted: Cox Proportional Hazards Model


• Matching

• Baseline Analysis for Matched

Cohorts

• Outcome Variables

• Outcome Analysis

Results

Variables CS Advice NRT p

n 40,799 17,121

Age at IPD (years) Mean (SD) 47.9 (11.6) 46.8 (11.3) <0.001

Height (m) Mean (SD) 1.7 (0.1) 1.7 (0.1) <0.001

Weight (kg) Mean (SD) 76.7 (17.9) 76.0 (17.8) <0.001

BMI (kg/m2) Mean (SD) 26.7 (5.5) 26.6 (5.6) 0.178

Gender (Male) n (%) 18,776 (46.0) 8847 (51.7) <0.001

BMI Category: Underweight n (%) 854 (2.4) 407 (2.7)

0.101 Normal n (%) 14,456 (40.8) 6321 (41.5)

Overweight n (%) 12,169 (34.3) 5135 (33.7)

Obese n (%) 7989 (22.5) 3373 (22.1)

Year of IPD 2006 n (%) 18,005 (44.1) 10,136 (59.2)

<0.001 2007 n (%) 13,788 (33.8) 4901 (28.6)

2008 n (%) 9006 (22.1) 2084 (12.2)

Baseline Variables - demographics

Baseline Variables: co-morbidities

1 Read Code at any time, 2 At any time prior to and including IPD 3 Calculated using the Charlson Comorbidity Index over 1 year prior to & including IPD

0

4

8

12

16

CS Advice

NRT

*

*

*

*

* *

*

* *

% cohort

*p<0.05

Baseline Variables – therapies

0

2

4

6

8CS AdviceNRT%

cohort

*

*

*

*

*p<0.05


Systolic Blood Pressure Mean (SD) 2496 (6.1) 1034 (6.0) <0.001

Diastolic Blood Pressure Mean (SD) 866 (2.1) 421 (2.5) <0.001

GP Consultations Mean (SD) 52 (0.1) 26 (0.2) <0.001

Total GP Consultations 0 - 2 n (%) 10,280 (25.2) 3095 (18.1)

<0.001 3 - 5 n (%) 11,366 (27.9) 4489 (26.2)

6 - 10 n (%) 9972 (24.4) 4616 (27.0)

11+ n (%) 9181 (22.5) 4921 (28.7)

GP consultations for CHD 1+ n (%) 76 (0.2) 71 (0.4) <0.001

GP Consultations for CerebroVD 1+ n (%) 130 (0.3) 88 (0.5) <0.001

Total OPD Attendance 1+ n (%) 5641 (13.8) 2731 (16.0) <0.001

Baseline Variables – characteristics


• Matching


Cohorts



Results

• To reduce difference between cohorts, cohorts

populations were matched

• 2 SC Advice : 1 NRT

• Patients were matched on:

o Gender

o Diabetes

o Cardiovascular Disease

o Hypertension

Matching


• Matching


Cohorts



Results


n 33,852 16,926

Age at IPD (years) Mean (SD) 46.96 (11.2) 46.87 (11.3) <0.001

Height (m) Mean (SD) 1.69 (0.1) 1.69 (0.1) 0.839

Weight (kg) Mean (SD) 75.99 (17.9) 76.06 (17.8) 0.785

BMI (kg/m2) Mean (SD) 26.6 (5.6) 26.6 (5.6) 0.835

Gender (Male) n (%) 17,328 (51.2) 8664 (51.2) NA

BMI Category: Underweight n (%) 739 (2.5) 400 (2.7)

0.660 Normal n (%) 12,264 (41.6) 6245 (41.5)

Overweight n (%) 9923 (33.6) 5080 (33.7)

Obese n (%) 6574 (22.3) 3334 (22.1)

Year of IPD 2006 n (%) 14,920 (44.1) 10,019 (59.2)

<0.001 2007 n (%) 11,466 (33.9) 4839 (28.6)

2008 n (%) 7466 (22.1) 2068 (12.2)

Baseline Variables – demographics

Matched

Baseline Variables – co-morbidities

Matched

1 Read Code at any time, 2 At any time prior to and including IPD 3 Calculated using the Charlson Comorbidity Index over 1 year prior to & including IPD

0

4

8

12

16CS Advice

NRT

% matched cohort *

*p<0.05

* * * *

Baseline Variables – therapies

Matched

0

2

4

6

8CS Advice

NRT

% matched cohort

*p<0.05

* *

* *

*

*


Systolic Blood Pressure Mean (SD) 130.9 (18.7) 130.8 (18.1) 0.018

Diastolic Blood Pressure Mean (SD) 79.5 (11.0) 79.2 (10.6) 0.012

GP Consultations Mean (SD) 7.5 (7.5) 8.8 (8.6) <0.001

Total GP Consultations 0 - 2 n (%) 8352 (24.7) 3077 (18.2)

<0.001 3 - 5 n (%) 9486 (28.0) 4443 (26.2)

6 - 10 n (%) 8371 (24.7) 4565 (27.0)

11+ n (%) 7643 (22.6) 4841 (28.6)

GP consultations for CHD 64 (0.2) 70 (0.4) <0.001

GP Consultations for CerebroVD 117 (0.3) 87 (0.5) 0.004

Total OPD Attendance 4707 (13.9) 2689 (15.9) <0.001

Baseline Variables – characteristics

Matched


• Matching


Cohorts



Results

Outcomes – 52 week outcome

0

0.2

0.4

0.6

0.8

1

Coronary Heart DiseaseDiagnosis

Cerebrovascular DiseaseDiagnosis

All Cause Mortality

CS Advice

NRT

* *

% matched cohort

*p<0.05

*

0

2

4

6

8

10

12CS Advice

NRT

% matched cohort


*p<0.05

* *

*

*

*

*


Healthcare Utilisation Variables CS Advice NRT p

Total Primary & Secondary Care 1 n (%) Consultations for CHD or Cerebrovascular Disease 2+ n (%)

135 (0.4) 107 (0.6) 0.110

54 (0.2) 49 (0.3)

GP Consultations for CHD n (%) 89 (0.3) 74 (0.4) <0.001

GP Consultations for Cerebrovascular Disease n (%) 94 (0.3) 82 (0.5) 0.065

No significant differences in OPD attendances or hospitalisations for CHD or Cerebrovascular Disease


• Matching


Cohorts



Results

Secondary Outcomes – 52 week Adjusted for Baseline Confounders

Time to first Coronary Heart Disease diagnosis

Time to first CardioVD diagnosis (ex prior Hx)

Time to first Cerebrovascular disease diagnosis

Time to first CerebroVD diagnosis (ex prior Hx)

All Cause Mortality

All Cause Mortality (ex prior Hx)

Primary & Secondary Care Consultations for CVD

Primary & Secondary Care Consultations for CVD (ex prior Hx)


Cardiovascular Disease Cerebrovascular Disease






All Cause Mortality


Primary & Secondary Care Consultations for CVD

Primary & Secondary Care Consultations for CVD (ex prior Hx)


Mortality






All Cause Mortality


Consultations for CHD & CVD

Consultations for CVD (ex prior Hx)



Time to first CHD diagnosis (ex prior Hx)

Time to first Cerebrovascular Disease diagnosis

Time to first CVD diagnosis (ex prior Hx)

All Cause Mortality


Consultations for CVD

Consultations for CVD (ex prior Hx)

Deaths in 52 week Outcome Period

• Why?

o Predisposing factors?

– Demographics

– Co-morbidities

– Therapies

– Other?

Baseline Characteristics – Deaths in 52 week

Clinical Outcomes

0

20

40

60

80

200

6

200

7

200

8

CO

PD

Angin

a

CC

I S

core

= 1

CC

I S

core

= 2

Beta

-Blo

cker

Antipla

tele

t

Year of IPD Comorbidities Therapies

CS Advice

NRT%

cohort

*p<0.05 *

* *

* * *

* * *

Baseline Characteristics – Deaths in 52 week

Healthcare Utilisation

0

10

20

30

40

50

60

70

0 - 2 3 - 5 6 - 10 11+

Total GP Consultations Total OPDAttendance

CS Advice

NRT% cohort

*p<0.05

* * *

*

*

Conclusions

• NRT is strongly associated with increased risk of

o coronary heart disease

o cerebral vascular disease

o all cause mortality

Increased Risk is independent of prior history

• Death in NRT cohort associated with

o Earlier formulations of NRT

o Higher prevalence of COPD but not Angina

o Prescription of Beta-blockers, anti-platelet therapies

Limitations

• Availability of NRT

o NRT available over the counter in UK

• Data Availability

o Limited to data held in GPRD

o Are we missing other confounding factors?

• Causal Link

o No information on causality

Team Effort

• Data Management Julie von Ziegenweidt et al

o Protocol Design

o GPRD to usable dataset

o Matching

• Statistics Annie Burden et al

o Protocol Design

o Statistical Analysis

• Research Team David Price & Erika Sims

o Protocol Design

o Manuscript

Questions

Creating your own database: examples

from practice iHARP - a 6 country

cross-sectional database

Stanley Musgrave

Senior Research Fellow

University of East Anglia

• 1. Why and What is “iHARP”?

• 2. How has iHARP been set up

o International steering committee

o Mix of CROs and local champions

• 3. What type of information is in the iHARP database

• 4. iHARP data collection - UK: application and use of

questionnaires

• 5. iHARP data collection – Other countries: the website

and use of questionnaires

• 6. Use of the database for research; CRITIKAL

• ??Lessons from iHARP for other databases ??

Outline…

• In the earlier presentations, a “Standard” OPC database has been presented. While it remains the core, it isn‟t the only thing possible.

• Study designs – “The Standard” (retrospective observational), and alternatives: cross-sectional observational and others

• Data sources – “the Standard” (OPCRD and GPRD), and alternatives: supplemental clinical data

• Clinical and geographic settings – “The Standard” (UK general practice) and alternatives:

o UK, Europe and Australia;

o Pulmonologists, GPs, nurses, Respiratory technicians, pharmacists, in their clinical settings

…a quick word about issues that

influence what’s in your database….

• Implementation of the IPCRG’s Helping Asthma

in Real-world Patients (HARP) Strategy (i-HARP)

• A respiratory review service

• A multi-country collaboration

The Background

What and Why is iHARP?

From the “Original” Harp programme

Promote an integrated approach to asthma

management, which focuses on doctor and

patient behaviour and sharing of best

practices

All aspects of real world inhaler efficacy:

handling, airflow, device type, and particle

size, as well as compliance, individual

phenotype factors

Understanding the reasons for poor

asthma control

Haughney J, et al. Respir Med 2008; 102:1681–1693.

Incorrect

diagnosis

Asthma control assessed

Uncontrolled: either current

symptoms or exacerbations

Well controlled

Continue/consider

step-down

Poor

compliance Rhinitis Smoking

Inadequate

or

incorrect

therapy

Other

phenotypes

Virus-

associated

wheeze

Exercise

induced

Low necessity High concerns Mixed devices Increased

inflammation

Produces

steroid

resistance

Not right for

that patient

Need for

more therapy Side effects

Concerns

Intrusiveness

Poor training

Erosion

Smoking Poor

inhaler

technique

Errors with the inhalation manoeuvre

• Dose preparation

• Do not exhale

• MDI – inhalation time is too short so flow is too fast

• DPI – inhale too slow – not enough acceleration

Percentage of patients making one error and the perception of their GPs

Molimard et al J Aerosol Med 2003; 16: 249-254

Perc

enta

ge

Aerolizer Autohaler Diskus pMDI Turbuhaler n= 769 728 894 552 868

Patients making at least 1 error

GP opinion – patient inhaled the right dose

• 30% of patients do not exhale before their

inhalation Molimard et al J Aerosol

Med 2003; 16: 249-254.

• Greater relative lung deposition when exhaling

to RV Hindle et al. Thorax 1993; 48 :607-610

• Alveolar deposition ↑40% for each ↑1L of

inhaled volume Pavia et al. Thorax

1977.; 32: 194-197.

LUNG VOLUME and LUNG DEPOSITION

Inhalation Profiles Asthmatic Adults when using a DPI

0

20

40

60

80

100

120

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

Time (sec)

Inh

alat

ion

Flo

w R

ate

(l/

min

)

PIF: 101-108 L/min

PIF: 58-62 L/min

2.32 (85%)

1.64L (72%)

1.36L (35%)

1.35L (125%)

2.00L (55%)

1.14L (95%)

Inhaled volume (% pred FEV1)

DPI OPERATING PRINCIPLE

Energy Input Energy Carrier Process Result

Inspiratory Force

Flow & Pressure

Deaggregation Fine Particle

Mass

Lactose

Drug

Airflow Generated By Patient’s Inspiratory Effort

Turbulent Energy R Q P =

Integrated enhanced asthma reviews in Real-world Patients

• Patients invited for iHARP Review are:

• age 18+

• Have a current asthma diagnosis/therapy

• Are at BTS/SIGN Asthma step 3 or 4

• Patient filled questionnaire, clinician observation of inhaler use, and inhalation assessment (Spirotrac /AIMS)

• Inhaler handling errors - a priori defined critical errors, • with planned work to refine and examine several items considered likely to be

critical - potentially critical

A multi-country collaboration

◦ UK,

◦ Netherlands, France, Norway, Sweden, Italy, Spain, Australia

So, What is iHARP?

How has iHARP been setup?

iHARP steering committee

• UK: David Price, Henry Chrystyn, John Haughney, Dermot

Ryan, Kevin Gruffydd-Jones

• France: Nicolas Roche, David Costa

• Italy: Federico Lavorini, Alberto Papi, Antonio Infantino

• Spain: Miguel Román Rodríguez

• Germany: J Christian Virchow

• Sweden: Karin Lisspers, Björn Ställberg

• Australia: Sinthia Bosnic-Anticevich

• Norway: Svein Henrichsen

• Netherlands: Thys van der Molen

Country lead Service implementation Practices

Recruited

Spain Miguel Roman Spanish Primary Care Respiratory Group GP network (GRAP).

19

France Nicolas Roche

Managed by College Régional des Généralistes Enseignants in Auvergne. GPs recruited by David Costa and GP network

Pending

Netherlands Thys van der Molen Ellen van Heijst & asthma COPD service network Reviews

starting

Italy Alberto Papi Andrea Romangini & Ricerche Nouveau (CRO), supported by Federico Lavorini and Antonio Infantino’s network of GPs

5 (to increase

to 20 in July)

Spain Miguel Roman GRAP GP network 19

Sweden Bjorn Stallberg Karin Lisspers & network of GPs 1

Norway Svein Hoegh Henrichsen

Beraki Ghezai and network of GPs. July start

Australia Sinthia Bosnic-Anticevich

Implemented in community pharmacies 46

Germany J Christian Virchow - -

iHARP dataset

• Unique international dataset containing: o Clinical records

o Patient reported outcomes

o Clinician reviews

• Comprehensive data on treatment, patient reported outcomes, adherence, smoking status etc.

• Personalised feedback for both patients and healthcare professionals

• Customisation to national standards of practice

• Anonymous database for research purposes (such as the CRITIKAL study)

iHARP in UK: Streamlined process

Patient Questionnaire

Consultation with respiratory clinician

and patient

Patient Feedback -review -therapy

-risk identification -inhaler technique

Review of routine practise data

Patient fills in online

Questionnaire

Review of routine practise data Patient Information

in database

iHARP UK implementation

o Current medication

o Effects of asthma

symptoms on daily life

o Confounding factors

(e.g. smoking, rhinitis)

o Side effects

o Adherence to

medication and

attitudes about

adherence

Questionnaire

- online or in paper form, covering:

Page 1 of 2 Page 2 of 2

iHARP review: internet technology

patient-reported information

Inhalation assessment

Spirotrac sub-group

Inhalation time ≥3 sec for MDI is acceptable; <3 sec is critical

Objective acceleration assessment:

MDI PIF L/min

DPI low resistance IF L/min at 0.4 sec

DPI high resistance IF L/min at 0.4 sec

Acceptable 30 to 90 60+ 90+

potentially critical

<60 <90

Critical <30, or >90 <30 <60

MDI Does not remove Cap Critical Puff 1 - Does not shake before actuation Error Puff 1 - Does not breathe out Potentially Critical Puff 1 - Exhalation into the inhaler Error Puff 1 - Does not hold inhaler upright Critical Puff 1 - Puts inhaler in mouth but does not seal lips Potentially Critical Puff 1 - Does not have head tilted such that chin is slightly upwards Error Puff 1 - Actuation not corresponding to inhalation; actuation before inhalation Critical Puff 1 - Actuation not corresponding to inhalation; actuation is too late Critical Puff 1 - Inhalation is not slow and deep - defined as lasting at least 3 seconds Potentially critical Puff 1 - Failure to actuate Critical Puff 1 - Failure to inhale Critical Puff 1 - Inhalation through the nose Critical Puff 1 - No breath-hold (or for less than 3 seconds) Error Patient coughed during the inhalation data clarification Second dose within 30 seconds Error No Repeat second inhalation Potentially Critical items listed as "Puff 1" repeated for "Puff 2" Patient coughed during the inhalation data clarification After second inhalation - doesn't replace cap Error When asked - patient does not know how to tell that their device is empty Critical Patient has an expired device Potentially critical If on Fostair, ask if they know how long they can use their inhaler after receiving it from the pharmacy - should be less than 20 weeks/5 months.) potentially Critical Patient did not bring their own device to the clinical visit data clarification Does not mention priming when asked: "What do you do when you haven't used your inhaler for 24 hours? (Evohaler 1 week, Fostair 2 weeks)" Potentially Critical Does not mention priming when asked: "What do you do when you use your inhaler for the first time?" Potentially Critical

Feedback: considerations

for therapy and

management, to

the clinician,

who confirms its

appropriateness,

and

communicates to

patient.

Anonymous ID

saved locally

• Patients invited for iHARP Review are:

• age 18+

• Have a current asthma diagnosis/therapy

• Are at BTS/SIGN Asthma step 3 or 4

• Implementation in a variety of clinical settings

• GPs, referral assessment clinics

• Questionnaire, clinician review, handling error

observation & inhalation assessment with AIMS

iHARP “non-UK” implementation

iHARP internationally

Differences in the international process

• Clinical Implementation

o Nurses, GPs and pulmonologists

• Consultation setting

o Referral clinic, primary care site

• Airflow assessment

o AIMS2 rather than spirometry

• iHARP for research

o In the UK iHARP is a service provider; international ethics

approval allows the work to be used for research purposes

iHARP Global website

The Vitalograph Aerosol Inhalation Monitor 2 (AIM 2) is a small desktop device powered by mains batteries and designed to objectively assess patient’s inhaler technique. It has a placebo pMDI and DPI attached and utilizes a system of visual indicator lights to assist someone to assess their own or observe another persons’ ability to use an inhaler.

Flow indicators: Green indicator light confirms inhaler actuated within 3 seconds after starting to breathe in & while inhaling at an inspiratory flow between 10-50 l/min.

Syncronisation indicator: Green indicator light confirms coordination between pMDi

actuation and inspiration is present.

Breath Hold indicator: Acceptable breath-holding times indicated by green light;

Three indicator lights: confirming correct (green) or incorrect (red) technique:

• A different model

• Labnoord referral clinical laboratory, seeing patients

referred by GPs

• Questionnaires – a mix of existing Labnoord form

and a supplemental form for OPCRD,with clinical

assessment.

The Netherlands…

LabNoord Standard iHARP supplemental

Handling errors and AIMS forms and assessment

+ LabNoord routine clinical session data is collected as routine

LabNoord database subset

1. A LabNoord clinic session invitation is sent to patient with these questionnaires:

2. In the LabNoord clinic

Copy, Courier to RIRL; For scanning to Database

Within OPCRD, an iHARP “Netherlands subset” Data available for future use

(Anonymised ... with Unique ID) transfered to OPCRD

(with the Unique ID)

Data entry

• Aim:

o To identify the prognostic (patient- and treatment-related) factors

associated with asthma control in patients receiving maintenance

fixed-dose combination (FDC) inhaled corticosteroid / long-acting

beta2-agonist (ICS/LABA) therapy in primary care.

• Exposures -Patients receiving any of the following inhaled therapies:

– Fluticasone / salmeterol (FP/SAL; Seretide®) via Diskus DPI

– FP/SAL (Seretide®) via MDI

– Budesonide / formoterol (BUD/FOR; Symbicort®) via Turbuhaler (including patients on SMART-use as maintenance and reliever-therapy)

– Beclometasone dipropionate/ formoterol (BDP/FOR; Fostair®) via MDI

Critikal

• UK: David Price, Henry Chrystyn, John Haughney, Dermot

Ryan, Kevin Gruffydd-Jones

• France: Nicolas Roche, David Costa

• Italy: Federico Lavorini, Alberto Papi, Antonio Infantino

• Spain: Miguel Román Rodríguez

• Germany: J Christian Virchow

• Sweden: Karin Lisspers, Björn Ställberg

• Australia: Sinthia Bosnic-Anticevich

• Norway: Svein Henrichsen

• Netherlands: Thys van der Molen

Critikal – the iHARP steering Committee

Country Number of patients

UK and Australia 2500

France

2500

Italy

Germany

Spain

Sweden

Norway

TOTAL 5000

Critikal

(1) Asthma control: ATAQ, GINA

(2) Risk assessment - exacerbations in the prior year

(3) Inhaler technique:

(a) Subjective patient perception.

(b) Subjective acceleration assessment

(c) Objective clinician technique assessment

(Sub-groups only) (d) Objective acceleration assessment

(4) Adherence

(a) Subjective adherence assessment: patient perception

(b) Objective adherence assessment: prescription refills

…

Baseline measurements

• Supported by:

o Research in Real Life

o Grant from Mundipharma

Critikal

• Your research question and protocol

• Selecting and specifying data variables from the

OPCRD

• Time periods, selection and eligibility criteria

• Baseline description, demographics, risk factors…

• Other issues (matching…)

•

So. Creating your own database…

Designing your own project

Practical Group Work

Publishing Real-life Research


• Displaying The Data

o Statistics

o Study Limitations

• Research Publication

o Real-life research becoming more accepted

o Not just accepted, but asked for!

Challenges of Real-life publications

Displaying real-life research: reviewers are

used to RCT format

RCT Real-life research

A: Initiation or step-up of therapy B: Switching therapy

Kenneth Stanley Circulation. 2007;115:1164-1169

Adjusting for confounding: minimising

baseline differences

• Convincing reviewers of the rigour of real-life

research

• Methods need to emphasise that confounding

factors have been minimised

o Statistical adjustments for baseline differences

o Patient matching

o Adjustments to address residual confounding

• RIRL definitions differ from clinical trial definitions

o Proxy Asthma Control

o Adherence

o Etc.

• Important to ensure that they are still clinically valid

• Current working on definition validation: showing the

validity, and importance, of our outcomes

Validation of Definitions

• ADEPT

o Independent committee that reviews OPC research

studies

o External review of OPC projects

• All trials registered on clinicaltrials.gov – results in

the public sphere

Approval and Validation

• Displaying The Data

o Statistics

o Study Limitations

• Research Publication

o Real-life research becoming more accepted

o Not just accepted, but asked for!

Challenges of Real-life publications

Who said this to The UK Royal College of Physicians?

“Randomised controlled trials (RCTs), long regarded at the 'gold standard‘ of evidence, have been put on an

undeserved pedestal.”

“They should be replaced by a diversity of approaches that involve

analysing the totality of the evidence-base.” Sir Michael Rawlins

Chairman of NICE

National Institute of Health & Clinical Excellence

Providing quality standards for healthcare

Classical RCTs: The Parachute Paradigm

Parachutes reduce the risk of injury after gravitational challenge, but their effectiveness has not been proved with randomised controlled trials.

(Smith GC, Pell JP. BMJ 2003; 327:1459-1461)

Common reviewer objections: RIRL solutions

• The possibility of residual confounding, eg, more knowledgeable / better doctors prescribing Qvar vs. the comparator o Possibilities listed in each study and specific solutions for

each study explored

• Outcome meanings differ from accepted clinical meaning (e.g “asthma control”) o Meanings well defined in paper and validation work

ongoing to justify them

• Confusion about how matching and statistical adjustments work o Full description detailed in methods section

• Patients with COPD are included in asthma studies o Smokers over 60 are excluded and very few diagnosis of

COPD younger than this

• Pragmatic trials designed and conducted to answer important questions facing patients, clinicians, and policymakers.

• Compare medical interventions that are directly relevant to clinical care or health care delivery and strive to assess effectiveness in real-world practice.

• Use broad eligibility criteria to ensure inclusion of patients whose care will be influenced by the trial‟s results.

Journals are starting to see the value of

real-life research

A treatment must be able to show its success rate in the “real-world”

In the USA: Interest is growing in “comparative

effectiveness research”

„„research evaluating and comparing health outcomes and the clinical

effectiveness, risks, and benefits of 2 or more health care interventions,

protocols for treatment, care management, and delivery, procedures,

medical devices, diagnostic tools, pharmaceuticals (including drugs and

biological agents), integrative health practices, and any other strategies or

items being used in the treatment, management, and diagnosis of, or

prevention of illness or injury in, individuals.‟‟

(J Allergy Clin Immunol 2011;127:123-7.)

CER as defined by the Patient Protection and Affordable Care Act

RIRL Research being published across a wide

variety of journals

• NEJM

• J Allergy Clin. Immunol

• CHEST

• J Asthma Allergy

• Respiratory Medicine

• Health Technol. Assess.

• PCRJ

Evidence from many designs: gives a full

picture of value of treatment

Evidence

Theoretical

Theoretical model provide

rationale

Classical double-blind

double-dummy RCTs

Gold standard, large range of

outcomes.

But not “real-life” patients,

compliance and represent <10%

of patients

Pragmatic trials

More real-life Broader inclusion

criteria Allow normal factors to

occur Usually randomised

Simple outcomes But still consent

& rigorous

Observational Data

Real-life patients Not randomised

Routine data Normal decisions Difficult to ensure

group comparability

Matching of case controls,

adjustment

Real-Life Research Seminar 25June12

Documents

Transcript of Real-Life Research Seminar 25June12