Epidemiology of multiple sclerosis: results from a large observational study in the UK

We conducted a population-based observational study using data from the UK’s Clinical Practice Research Datalink (CPRD), formerly the General Practice Research Database (GPRD). The CPRD contains prospectively collected data on all aspects of medical care for around 8 million people in the UK. Validation studies have shown the quality of the CPRD data, which have been used for several MS studies [3‐10], to be generally high [11‐14]. Also available are paper records that describe referrals, hospitalizations and other relevant medical history, often covering the patient from a young age. These provide an ideal resource to validate MS diagnoses and determine dates of MS onset. Details of the database have been described elsewhere [11‐14]. This study was reviewed and approved by the Independent Scientific Advisory Committee for UK Medicines and Healthcare products Regulatory Agency database research.

We identified all patients in the CPRD who had a first diagnosis code (Online Resource 1) for MS between 1st January 2001 and 31st December 2006 who had at least 2 years of information in their electronic records. Cases were identified only through 2006 to provide sufficient follow-up to develop new outcomes, since the primary objective of this research was to study the prognosis of people with MS. Among all potential cases we classified each according to the likelihood that the MS diagnosis was valid according to the 2005 revisions to the McDonald MS criteria where data were available [3, 15]. This process resulted in a series of cases validated through detailed paper and electronic records, or electronic records only. For patients with paper records, we were able to get additional information on the type of MS [relapsing–remitting MS (RRMS), primary progressive MS (PPMS), secondary progressive MS (SPMS), and unknown type] and the presence of MS symptoms at onset. We classified symptoms, where available, into four main groups: sensory, motor, optic neuritis, and other symptoms (including other optic and visual anomalies, other cranial nerve anomalies, dysarthria, and asthenia). For those with no symptoms recorded, we treated them as a separate group with missing symptoms. In addition, all incident and prevalent MS cases still present in the 2012 data update, identified and validated in our earlier studies conducted using the GPRD (1993 and 2000) [5‐10] were added to this 2001–2006 case set. Further details of the study design and validation process have been published previously [3].

Information on the following covariates was extracted from the database: age at first MS diagnosis, sex, and lifestyle factors closest to and before the first MS diagnosis including smoking, body mass index and alcohol abuse (Table 1). Information on acute infections (recorded on or after first MS diagnosis), chronic comorbidities (recorded at any time in the database), and MS treatment (recorded on or after first MS diagnosis) were also extracted. Comorbidities and MS treatment evaluated are listed in Table 2. MS treatment included disease-modifying medications (interferons, glatiramer acetate, mitoxantrone, and natalizumab), steroids, and symptomatic management drugs (tizanidine, baclofen, amantadine HCL, modafinil, and fampridine). We were unable to obtain complete information on use of disease-modifying medications because, in the UK, they are generally prescribed in secondary care and are not always captured in general practice (GP) records.

Patient characteristics were described and predictive factors for all-cause mortality were first identified by calculating crude hazard ratios (HRs) with 95 % confidence intervals (CIs) using univariate Cox proportional regression analyses. Because death was the outcome of interest, information on acute infections closest to the time of death was captured. We identified the last instances of each of the acute infections before end of follow-up and used that event date to divide follow-up time into two periods: one not exposed to acute infection and one exposed to acute infection. If chronic comorbidities were recorded on or before the first MS diagnosis, follow-up time was treated as exposed to chronic comorbidities. If chronic comorbidities developed during follow-up, we used the first event date to divide follow-up time into two periods: one not exposed to chronic comorbidities and one exposed to chronic comorbidities. MS treatment was treated as time-varying in the Cox regression models where total follow-up time was divided into one-year periods and we identified whether these medications were prescribed at least once in each period. Given that age is a known risk factor for mortality and also an important confounder of the associations between the studied chronic comorbidities (such as heart disease, cancer, etc.) and mortality, we used age as the time-scale in the Cox regression models, where patients entered the analysis at the age at first diagnosis of MS and exited at their death or censoring age, whichever was earlier. We also estimated the adjusted HRs and 95 % CIs for each predictive factor in the Cox proportional models which included all predictive factors as well as year of birth. Because the proportional hazards assumption may not be met across age, we divided the data into one-year age strata for each MS patient and fit models separately within each age stratum. Since age is the time scale, these age strata refer to attained age, rather than to age at cohort entry. All analyses were conducted using SAS version 9.3 (SAS Institute, Cary, NC).

We identified 1879 people with a first diagnosis of MS between January 2001 and December 2006. Of these, 998 (53.1 %) had available paper records. These 998 MS cases were reviewed by a neurologist (G.J.F). Of these, 683 (69 %) were considered to have MS, 87 (9 %) as possible MS and the remaining 228 (23 %) as not MS, yielding a positive predictive value of 68 % (95 % CI 65.5–71.3). Out of 881 cases with unavailable paper records, validation based solely on electronic records yielded 429 probable and 229 possible MS cases (658 cases in total). Altogether 1278 incident and 63 prevalent MS cases were identified from 2001 to 2006 (Fig. 1 for MS validation process). In addition, 435 incident and 46 prevalent MS cases identified between 1993 and 2000 who were still present in the current database were retrieved from previous studies [5‐10], giving a total of 1822 MS cases (1507 definite or probable and 315 possible) including 1713 incident and 109 prevalent MS cases (Fig. 1). Because reliable information on MS type and symptoms at onset was only available for incident cases, we restricted the study analyses to the 1713 incident MS cases.

Sixty per cent of the 1713 incident MS cases were first diagnosed between ages 30 and 49 (data not shown); mean 42 years, and a majority were female (74 %). Current or former smoking and alcohol abuse were more prevalent among the MS patients who died during follow-up than those who did not die (p < 0.01 for all). Among patients whose diagnosis was made based on original clinical records (n = 851), 77 % had RRMS, 14 % had PPMS, 7 % had SPMS and 2 % had an unknown subtype; sensory, motor symptoms, and optic neuritis were the most common symptoms recorded at MS diagnosis (Table 1 shows further data on characteristics of the patient population at first MS diagnosis).

MS patients had a high prevalence of several concomitant conditions and MS treatment (Table 2). Infections were the most frequently recorded acute comorbidity on or after first MS diagnosis, occurring in 80 % of patients. Depression, recorded at any time in the database, was the most frequently recorded chronic comorbidity, occurring in 46 % of patients. Prevalence was higher in MS patients who died during follow-up than those who did not die: 25 vs. 7 %, p < 0.0001 for pneumonia and influenza; 52 vs. 31 %, p < 0.0001 for urinary tract infection; 12 vs. 5 %, p = 0.0009 for diabetes; 16 vs. 3.0 %, p < 0.0001 for heart disease; and 23 vs. 5 %, p < 0.0001 for cancer.

A total of 115 (6.7 %) MS patients died during 13,573 person-years of follow-up. The mean age at first MS diagnosis was higher in patients who died (51.4) than those who did not die during study follow-up (41.1). Mean age at death was 56.9. Current and former smokers at the time of diagnosis had increased risks of death (adjusted HR 2.0, 95 % CI 1.2–3.4; and adjusted HR 2.5, 95 % CI 1.3–4.5, separately), as well as patients with a history of alcohol abuse at the time of diagnosis (adjusted HR 7.6, 95 % CI 3.2–17.7). Among MS cases confirmed by original clinical records, patients with PPMS had a higher risk of death compared with patients with RRMS (adjusted HR 2.2, 95 % CI 0.9–5.5), although it was borderline statistically significant. Compared with patients with sensory symptoms only, patients with other types of symptoms had increased risks of death, particularly patients with multiple symptoms at onset who had a greatly increased risk (adjusted HR 9.0, 95 % CI 2.2–36.3). These effects were based on the 851 MS cases who had paper records and, therefore, details of first symptoms. We only had data on symptoms for 49 MS patients who died. Thus, this effect was based on a small number of cases, never-the-less, the finding was robust and was stronger when covariates were added to the model; see Table 3 for details. Several comorbidities were found to be significant predictors of death; adjusted HRs (95 % CIs) were: 2.7 (1.6–4.5) for pneumonia and influenza, 4.1 (2.7–6.3) for urinary tract infections, 2.2 (1.2–4.2) for heart disease, and 4.9 (2.9–8.0) for cancer. MS treatment was associated with a reduced risk of death in these data (adjusted HR 0.6, 95 % CI 0.4–1.0). See Table 4.