The Norfolk Arthritis Register Alan Silman arc Epidemiology Unit University of Manchester UK
Manchester
The Norfolk Arthritis Register A primary care based inception cohort study of patients with inflammatory polyarthritis
Norfolk
Why Norfolk? Geographically ‘isolated’ Geographically ‘isolated’ Stable population Stable population Single central major hospital Single central major hospital Excellent links primary to secondary care Excellent links primary to secondary care Local enthusiasm Local enthusiasm
Topics The NOAR methodology Key results –Classification of RA –Environmental risk factors –Outcome –Predictors of outcome –Treatment effects
Manchester Norwich NOAR : Recruitment Entry criteria -age > 16 years -registered with local GP -swelling of > 2 joints -duration > 4 weeks -onset since 1/1/90 Metrology assessment Apply ACR criteria
Metrologist Assessment
Data Collected
The Norfolk Arthritis Register (NOAR) To establish the incidence of IP and subset with RA To identify risk factors for the development of IP and RA To study the natural history of treated IP and RA To identify predictors of outcome in IP and RA Initial aims
The Norfolk Arthritis Register (NOAR) To investigate the epidemiology of cardiovascular disease in patients with IP (risk factors, incidence and outcome) To identify predictors of treatment response and non-response Current Major Aims
Key results Incidence of IP and RA
Estimates of the incidence of RA: Application of ACR criteria Incidence rate per 100,00 FemalesMales ACR criteria applied at baselineACR criteria applied over 5 years Age
Issues of Classification IP vs RA
Concept Early IP Recovery Another disease Established RA
Concept Early IP Recovery Another disease ? Treatment Established RA
Does early RA exist?
Are there differences between IP destined to differentiate into RA and other ‘causes’ of IP? Are there differences between IP destined to differentiate into RA and other ‘causes’ of IP?
Immunisation X ParvovirusX PsoriasisX Can we distinguish early RA from other forms of early arthritis?
Leiden model: prediction of outcome Goal: To discriminate at first visit between patients who will go on to have: self-limiting arthritis self-limiting arthritis persistent non-erosive arthritis persistent non-erosive arthritis persistent erosive arthritis persistent erosive arthritis
Leiden model: 7 variables Symptom duration at presentation Symptom duration at presentation Morning stiffness > 1 hour Morning stiffness > 1 hour Arthritis of > 3 joints Arthritis of > 3 joints Bilateral compression pain of MTPs Bilateral compression pain of MTPs Rheumatoid factor Rheumatoid factor Anti-cyclic citrullinated peptide antibody Anti-cyclic citrullinated peptide antibody Erosions in hands or feet Erosions in hands or feet
Validation of Leiden model erosive vs non-erosive arthritis In presence of persistence In presence of persistence Radiological criterion omitted Radiological criterion omitted LeidenNOAR (n= 526)(n=486) Prediction model ROC ACR criteria ROC
Key results Risk factors for the development of IP and RA
Sources of Data Descriptive Analysis
Local Clustering of RA Silman et al., 1999
Jan 1990 June Jan 1991 June Jan 1992 June Month of onset Number of new cases All cases UIP RA Onset of Disease by Month Silman et al., 1997
Observed & Expected Events in Relation to Time & Distance Silman et al., e7 -1e D Time Distance
Socioeconomic Deprivation vs RA Incidence by census ward Bankhead et al., J Rheum 1996 Indicator rsrsrsrs Households in rented accommodation Overcrowded accommodation Householders with no CH Households with no access to a car Male unemployment (age ) -0.03
Socioeconomic Deprivation & RA Bankhead et al., J Rheum 1996 Social Class Incidence /100,000 * IV & V combined for men *
Sources of data Case control studies Case control studies 1. Internal NOAR Cases 1992 (n=165) : –aged –symptom duration < 12 months Controls: 2 per case from referring primary care
Lifestyle Factors SmokingObesity
Association of Smoking with Severe RA: Rheumatoid Nodules CurrentExNever Harrison et al., Arth Rheum 2003 Odds Ratio (95% CI)
Hormonal Risk Factors TerminationOral ContraceptiveMiscarriage
Symmons et al., 1997 CasesControls % Association between Prior Blood Transfusion and RA
2. NOAR EPIC Link
Co-occurrence of NOAR & EPIC in same population Area for new cases of IP referred to NOAR EPIC practices
European Prospective study of the Incidence of Cancer (EPIC-Norfolk) Baseline assessments Random sample (n= 25,000) 45 – 75 years Recruited 1993 – 1997 Health and lifestyle questionnaire Height and weight
Prospective ‘nested’ case control study Free of IP at baseline Free of IP at baseline Subsequent registration with NOAR Subsequent registration with NOAR 2 per case 2 per case Matched: Matched: - age (± 3 years) - gender - gender - within 3 months of baseline assessment baseline assessment 73 Cases Controls
EPIC Diet Survey 7 day detailed food diary with portion sizes 7 day detailed food diary with portion sizes
Fruit Intake (g) and Development of IP Highest (ref)MiddleLowest * Adjusted for energy intake, smoking, red meat intake Odds Ratio (95% CI)* Pattison et al., ARD 2004 Tertile
Tertiles of Vitamin C Intake (mg) Highest (ref)MiddleLowest * Adjusted for energy intake, smoking, protein intake Pattison et al., ARD 2004 Odds Ratio (95% CI)* Tertile
Tertiles of -cryptoxanthin Intake (µg) HighestMiddleLowest (ref) Odds Ratio (95% CI)* * Adjusted for energy intake, smoking, protein intake Pattison et al Tertile
HighestMiddleLowest (ref) RA and Dietary Zeaxanthin Intake Odds Ratio (95% CI)* * Adjusted for energy intake, protein, smoking Pattison et al 2005 Tertile
Red Meat & Meat Products and Development of IP * Adjusted for energy intake, smoking, fruit intake MiddleLowest (ref)Highest Pattison et al., A & R 2004 Odds Ratio (95% CI)*
Are the Diet Effects Independent? Vitamin C mg/day HighMiddleLow Red Meat g/day HighMiddleLow Odds Ratio (95% CI)*
Key results The natural history of treated IP and RA
Outcomes investigated PersistenceRadiological damage Physical function (HAQ)Economic costs Health status (SF-36)Co-morbidity Work disabilityMortality
Work disability Year of onset
All cause mortality Seropositive patients Men Women Norfolk 0 1 SMR Inflammatory polyarthritis SMR MenWomen Norfolk SMR = 1.13 SMR = SMR =1.51 SMR = 1.41
Cardiovascular mortality: Influence of RF Status Males Females RF- SMR (95% CI ) RF+
Key results Predictors of outcome geneticenvironmentaltreatment
X-ray strategy NOAR Time from registration Patients X-rayed None 3 ACR criteria at baseline 2 ACR criteria at year one 2 ACR criteria at year two and no erosions on any previous X-rays All patients
Timing of first erosions Risk setTime of 1st “erosion free” X-ray Median (IQR) Timing of 2nd X-ray Median (IQR) % erosive at 2nd X-ray Incidence rate of 1st erosions (per 1000 pm) (95% CI) (16-20) 29 (26-31) 41 (37-45) 18 (16-22) 66 (64-69) 69 (66-73) 75 (70-84) (21-29) 5(4-8) 7(5-10) 13(9-19)
NOAR: Predicting radiological erosions Risk group RF > 40 Initial duration > 3 months Probability of erosions XXXX XXXX Overall performance: PPV 61% NPV 74%
Role of genetic factors HLA.DRBI HLA.DRBI Cytokine Cytokine –TNF –IL1 etc etc MMP MMP MBL MBL MIF MIF
Weak association with shared epitope, less strong than in clinic based studies Weak association with shared epitope, less strong than in clinic based studies Few candidates tested were predictors of presence/severity erosions Few candidates tested were predictors of presence/severity erosions Genetic Factors
? Confounding effect of therapy
Propensity models Bias in treatment assignments “Confounding by indication” Variable duration of exposure to treatment Problems Solution Assessing the effect of treatment
In observational studies : It is not random who will get DMARD therapy Treated patients have more severe disease Therefore ‘bias in allocation’ occurs Adjustment for this effect is needed
Propensity modelling Logistic model used to predict treatment decision Using disease characteristics that inform treatment decision Each individual given probability of being treated = propensity score
Distribution of HAQ scores at year 5 Never on DMARDs <6 months6-12 months>12 months Delay from symptom onset to start of first DMARD
Odds of moderate disability (HAQ 1.0) at 5 years Delay from onset to start of treatment
Odds of moderate disability (HAQ 1.0) at 5 years (Models include propensity scores & hospital referral) Delay from onset to start of treatment
Odds of moderate disability (HAQ 1.0) at 5 years Without propensity score With propensity score Delay from onset to start of treatment Never on DMARDs/steroids < 66-12> 12 Months
Larsen score at year 5 adjusted for propensity score Delay to start of first DMARD < 6 months6-12 months> 12 months No Treatment
Patients treated with DMARDS had worse disease at presentation and worse outcome The greatest benefit of treatment was seen in those treated within six months
Jt Principal Investigator :Deborah Symmons Research Fellows :Marwan Bukhari Beverley Harrison Nicola Goodson Research Assistants:Clare Bankhead Nicola Wiles Dorothy Pattison Statisticians: Paul Brennan Mark Lunt Research nurses, consultant rheumatologists Acknowledgements