Changes in the ascertainment of multiple sclerosis

The McDonald criteria are the most recently proposed diagnostic criteria for multiple sclerosis (MS) and rely heavily on the use of MRI.¹ Even prior to the introduction of these criteria several investigators reported a decreasing delay in the time between symptom onset and diagnosis of MS.^2–4 Retrospective application of the McDonald criteria leads to a diagnosis of MS within a year of symptom onset in three times as many patients presenting with clinically isolated syndromes as would purely clinical criteria.⁵

Earlier diagnosis of MS and changes in diagnostic criteria may affect incidence and prevalence rates, leading to temporal changes in disease rates.⁴ With formal alterations in diagnostic criteria, increased availability of MRI, and awareness of MS therapies, the proportion of minimally affected persons diagnosed with MS may be increasing. This may affect viability of proposals to use natural history or historical controls for testing new therapies, as well as leading to inaccurate power calculations for studies relying on traditional rates of disease progression in newly diagnosed patients with MS.^6,7 For clinicians, the meaning of treatment benefit may be altered or at least not be comparable to previous era trials. This potential could result from evidence suggesting that treatment earlier in the course of the disease produces larger effects.^8–10

Our objectives were to use a patient self-report registry 1) to assess temporal trends in the delay from symptom onset to diagnosis (DONDX) in MS and 2) to determine whether there are differences in the degree of disability at diagnosis by year of symptom onset.

Methods.

The NARCOMS Registry is a large patient self-report registry for MS and is a project of the Consortium of MS Centers (CMSC).¹¹ The registry has a current enrollment of 28,776 participants who registered between 1996 and 2004. For the purposes of this study we selected participants who reside in the United States, to limit heterogeneity due to different health care systems (n = 28,272); reported both age at symptom onset and age at diagnosis (n = 27,272); reported age at symptom onset as preceding age at diagnosis (n = 27,121); reported an age at symptom onset between 10 and 60 years, which excluded 1.7% of NARCOMS participants (fewer than 0.6% of MS patients have onset after age 60,¹² and fewer than 1% have onset before age 10¹³ [n = 26,648], thus the proportion of participants excluded is similar to the expected proportion of 1.6%; we excluded these participants because of concerns regarding potential heterogeneity with respect to diagnosis); and reported a year of symptom onset of 1980 or later so that MRI could contribute to the diagnosis. MRI was included in MS diagnostic criteria first in 1983¹⁴ (n = 16,581).

Demographic and clinical data were self-reported at enrollment, and missing data were not imputed. The data reported included age, sex, race, income, education, state of residence, age and year at first symptom onset, and year of MS diagnosis. Disability status was reported using two validated self-report measures: Performance Scales (PS) and Patient-Determined Disease Steps (PDDS).^15–17 We calculated the DONDX for each participant.

Results.

A total of 16,581 participants met the inclusion criteria for the study of the relationship between year of symptom onset and DONDX. Of these, 5,548 (33.5%) participants enrolled within 2 years of diagnosis. Characteristics of the study samples are shown in table E-1 (available on the Neurology Web site at www.neurology.org). Their characteristics are similar to those reported for the general MS population.^20,21 Participants who enrolled online were more likely to enroll within 2 years of diagnosis (p < 0.0001). At enrollment, more than 98% of the participants reported ever having an MRI, and 74.4% reported ever having a lumbar puncture. Because participants report only the date of their most recent MRI and lumbar puncture, we could not assess the changing role of these tests in the diagnostic process.

There was a steady decrease in DONDX with later year of symptom onset (p < 0.0001, Kruskal-Wallis test) (table 1). Figure 1 demonstrates that 50% of participants with initial symptom onset in 1980 to 1984 are diagnosed within 6 years, 50% of participants with symptom onset in 1990 to 1994 are diagnosed within 3 years, and 50% of participants with symptom onset after 2000 are diagnosed within less than a year. The Spearman correlation between year of onset and DONDX was r = –0.43 (p < 0.0001). The DONDX was shorter in men, African Americans, and those with later age at symptom onset between 1980 and 1999 (p < 0.0001) (see table 1). Participants who enrolled online had a slightly longer DONDX than those who enrolled via paper forms (5.20 ± 5.0 vs 4.08 ± 4.1 years).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Delay from symptom onset to diagnosis of multiple sclerosis by year of symptom onset.

Multivariate analysis showed that DONDX decreased with later year of symptom onset (table 2). The model was stratified by age at symptom onset, as that variable did not satisfy the proportional hazards assumption. Men had a shorter DODNX than women. There was an interaction between sex and year of symptom onset, so we ran separate models for men and women. This showed that the decrease in DODNX with later year of symptom onset was less marked for men than for women. Relative to African Americans, other races had a longer DONDX.

Disability.

The median severity of disability at enrollment was progressively less with later year of symptom onset whether reported using PS or using PDSS (p < 0.0001 for both). The proportion of participants with mild disability within 2 years of diagnosis progressively increased with later year of onset (p < 0.0001) (figure 2). With univariate analysis, a shorter DONDX (p < 0.0001), being female (p < 0.0001), being more educated (p < 0.0001), and having a higher income (p < 0.0001) were all associated with having milder disability based on the PDDS. African Americans had more disability at diagnosis (p = 0.006).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Disability by year of symptom onset: participants enrolling within 2 years of diagnosis.

In the polytomous logistic regression model, we initially did not include a term for DONDX because of the strong collinearity between DONDX and year of symptom onset (r = –0.90) in participants who enrolled in NARCOMS within 2 years of diagnosis. Later year of symptom onset was associated with decreased odds of having moderate (OR = 0.93 per year; 0.92 to 0.95) or severe (OR = 0.91 per year; 0.90 to 0.92) disability as compared to mild disability at diagnosis (see table E-2). Women and those with higher incomes and more education were also less likely to have moderate or severe disability at diagnosis. African Americans had increased odds of having moderate (OR = 1.59; 1.01 to 2.48) or severe (OR = 1.35; 0.78 to 2.35) disability. Later age at symptom onset was associated with increased odds of having moderate (OR = 1.06 per year; 1.05 to 1.07) or severe (OR = 1.03 per year; 1.02 to 1.04) disability.

Another model with the same covariates that also included a term for DONDX showed the same associations between year of symptom onset and disability, with decreased odds of having moderate (OR = 0.92 per year; 0.86 to 0.98) or severe (OR = 0.93 per year; 0.87 to 0.99) disability as compared to mild disability at diagnosis. Because participants with an early year of symptom onset and short DONDX were excluded, and some individuals with a later year of symptom onset will not be diagnosed yet, we stratified by DONDX. This produced similar results to those reported above.

We performed a series of additional analyses to assess the sensitivity of our results to the method of sample selection (data not shown). Results were similar when the disability analysis was restricted to the 3,526 participants who enrolled in NARCOMS within a year of diagnosis or expanded to include the 6,919 participants who enrolled within 3 years of diagnosis (data not shown). As the registry existed longer, a greater proportion of participants enrolled within a short time from diagnosis. Therefore, we stratified the analysis by year of enrollment in the registry and repeated the analyses. The ORs varied slightly but consistently showed that later year of symptom onset was associated with decreased risk of having moderate (range of OR: 0.85 to 0.94) or severe (range of OR: 0.89 to 0.97) disability at diagnosis. The results were also similar for the DONDX analysis with stratification by year of enrollment in the registry. When we restricted analysis to participants enrolled within 1, 2, or 3 years of diagnosis the hazard ratios were higher, and CIs were broader, but results were similar.

Discussion.

The diagnosis of MS has changed substantially since the first diagnostic criteria were developed. The most recently proposed diagnostic criteria (in 2001) for MS rely heavily on MRI, and have the potential to lead to more rapid diagnosis of MS in patients presenting with clinically isolated syndromes.⁵ We have shown that the delay from initial symptom onset to diagnosis of MS has steadily decreased since 1980. Men and African Americans had shorter delays from symptom onset to diagnosis, and were also more likely to have severe rather than mild disability at diagnosis. Older age at symptom onset was associated with greater disability. These findings are consistent with previous reports.^{2–4,22–26}

For epidemiologists, shortening of the delay between symptom onset and diagnosis has the potential to cause temporary increases in incidence and prevalence rates if this factor is not considered. The use of onset-adjusted disease rates has been proposed to reduce the impact of such changes in the delay from symptom onset to diagnosis on disease rates.²² Regression models can be used to describe the delay distribution and correct incidence rates.⁴ For patients these changes may mean earlier institution of therapy with the possibility of better long-term outcomes.^8–10 However, potentially larger effect sizes on such outcomes as relapse rates with earlier intervention may signal caution is necessary when comparing studies of therapeutic benefits.

Of potentially greater importance is our finding that in more recent years a greater proportion of diagnosed MS cases have a mild degree of disability even after accounting for DONDX. Given the emphasis on early treatment of MS, it is likely that more patients with milder disease will begin disease-modifying therapies. It is possible, even likely, that some of these mild cases might have a good prognosis even without treatment^27,28; thus patients may be subjected to unnecessary adverse effects and costs of therapy.²⁹ Unfortunately, the capacity to distinguish which patients will have a mild vs severe disease course is poor. Increased emphasis must be placed on identifying strong prognostic factors so that treatment may be targeted toward those who need it.

There are proposals to use natural history and historical controls as a means of reducing costs, and expediting evaluation of new MS therapies.⁶ Enrichment of treated cohorts with milder disease for comparison to natural history cohorts could lead to bias in favor of new therapies.³⁰ Sample size calculations for clinical trials may be inadequate if based on behavior of older, more severely affected cohorts of patients with MS. A recent clinical trial for primary progressive MS was halted early because of an unexpectedly low rate of progression in the placebo group.³¹

We can only speculate why more mild cases might be diagnosed. Two factors may contribute to this finding: availability of MRI has increased and awareness of disease-modifying therapies for MS is greater among physicians and the general public and this may lead to increased ascertainment, preferentially affecting mild cases that under previous circumstances would wait or remain undiagnosed. Rapid increases in incidence and prevalence of MS in central Finland have been attributed to improved ascertainment secondary to substantial increases in the use of MRI.² Changes in diagnostic criteria could also play a role. Severity of disease may be affected by genetic and environmental factors, and the prognosis for MS appears somewhat different depending on the population studied.^27,28,32 Thus it is also possible that the spectrum of MS disease severity may be truly changing due to changes in environmental factors, but it could also reflect the way in which the disease is manifested at various stages of severity.

There are potential biases to consider in interpreting our results. Participants in the NARCOMS Registry are volunteers, enroll at variable intervals following diagnosis, and provide disability information from the point of enrollment onward. Thus we did not use a random sample of patients with MS and the patients who responded to recruitment efforts for the registry may be somehow different than those who did not. All data are self-reported, and the accuracy of the diagnosis of MS is not verified against medical records. However, comparison with National Health Interview Survey (NHIS) data for MS shows that our patient population is similar to the general MS population with respect to demographic characteristics.²¹ Clinical characteristics are similar to those of the New York Multiple Sclerosis Consortium.²⁰

Some individuals with a more recent year of symptom onset (≥2000) and longer delays from symptom onset to diagnosis will not be diagnosed or enrolled in the registry yet, therefore the mean DONDX for persons with more recent symptom onset are likely underestimated. Patients with a short DONDX may be more likely to enroll in the registry if their symptom onset was recent. Among participants diagnosed within 4 years of symptom onset, 28.7% of those with onset between 1980 and 1984 were diagnosed within a year, while 73% with onset in 2000 were diagnosed within a year. This suggests that the mean DONDX is decreasing, although the magnitude of the effect may be overestimated since we do not have the remainder of the persons to be diagnosed with MS. Furthermore, this truncation bias would have to be substantial to be the source of our differences. If the median DONDX were 6 years as was found in 1980 to 1984, then for the median to drop to 3 years, assuming DONDX follows an exponential distribution, would require truncation of 40% of the cases for those with onset between 1990 and 1994. This amount of truncation seems highly unlikely given the fact that under the same exponential assumptions we would anticipate only 25% of the population were yet to be diagnosed 12 years after symptom onset.

We limited our analysis of the association between year of symptom onset and disability to participants who enrolled within 2 years of diagnosis in order to have data that reflect disability at the time of diagnosis. This means that individuals diagnosed prior to the existence of the registry who had a short delay from symptom onset to diagnosis were not included, thus truncating the sample. Some individuals with more recent symptom onset and longer delays from symptom onset to diagnosis will not be diagnosed yet and would not be included in our analyses. If individuals with more severe disease (i.e., greater disability, more frequent relapses) are diagnosed faster than more mildly affected individuals this would tend to enrich the study sample, but should not affect our conclusions. If individuals with more severe disease are diagnosed faster now than in the past, this would enrich the earlier year of symptom onset groups with mildly affected individuals and bias toward the null hypothesis. Stratification by DODNX did not materially affect our results.

Our results were robust to changes in our assumptions, thus we believe our findings to be sufficient to recommend further investigation of this question, ideally in untreated, population-based cohorts. The data did not allow us to assess the impact of MRI on changes in the delay to diagnosis or on disease severity, but this also deserves investigation.