Statistics from Altmetric.com
We write in relation to the article from colleagues at Beaumont Hospital, Dublin, who report that patients with amyotrophic lateral sclerosis (ALS) attending a multidisciplinary clinic had a better prognosis owing to better medical care than those attending general neurology clinics.1 In particular the survival of the bulbar onset patients was extended by 9.6 months if they attended the ALS clinic.
The inherent error in their conclusions relates to the manner in which the two populations compared were derived. Patients were recruited to the ALS clinic group up to one year after diagnosis; the general neurology clinic population was recruited immediately from the time of diagnosis. Patients who die from ALS within some months of the diagnosis are not available to the ALS clinic cohort but these are included in the general neurology clinic population. It is probable that patients living further from the ALS clinic and those who are more disabled by ALS attend their local neurologist. Thus the ALS clinic treated a group of fitter ALS patients, whereas the general neurologists saw all ALS patients regardless of medical, social, or economic factors. In their model the authors corrected for some factors predictive of a poor outcome in ALS which were all overrepresented in the general neurology clinic population (increased age, bulbar onset, and shorter duration of illness at presentation); they did not correct for a measure of baseline disease severity.
The effect of the recruitment bias can be seen in the survival graphs (figs 1–3) comparing the two populations. In the ALS clinic group there were no deaths in the bulbar onset group for 250 days, whereas in the general neurology ALS patients it seems that deaths occurred within 30 days (fig 2). It is difficult to be exact about the latter figures because no survival tables are given in the article. If the patients at risk (or the percentage alive) had been reported below the graphs, this would have been of use to the reader.2 It is noteworthy that the subsequent rate of decline in survival of the two groups is exactly the same; the only differences between the groups in all the measures are the initial 100% survival in the ALS clinic group for about 200 days (whole group and riluzole subgroup) and 250 days (bulbar onset subgroup). The reason for this initial survival advantage is that in order to attend the ALS clinic one must be relatively well and not requiring immediate hospital care. As the authors indicate in their discussion, even the ability to attend the ALS clinic on one occasion or one year after diagnosis conferred a significant survival advantage with a 25% reduction in mortality (p = 0.01).
The survival advantage of the ALS clinic group thus appears to reflect the increased mortality of the patients treated by general neurologists in the first 200–250 days; these patients are not well enough to attend the ALS clinic.
Inferior treatment by general neurologists is implied (for example, “less attention was paid to early introduction of gastrostomy feeding”—for which no evidence was produced) and is suggested as the reason for the increased mortality.
We accept that a multidisciplinary clinic is valuable in the management of ALS, but this paper is not scientific evidence for this view. The paper would have been useful if the authors had matched ALS clinic and general neurology clinic patients, even retrospectively, for age at onset, mode of onset, disability, and duration of illness. Patients should have been deemed to have entered the ALS clinic cohort only from the date of first attendance at the ALS clinic and not from the date of diagnosis, which may have been up to one year previously. A treatment effect of the ALS clinic can only be possible from the date of first attendance. Censoring early deaths in the clearly more ill general neurology cohort should also have been considered. By avoiding these biases one might have a possible estimate of the effect of attendance at the ALS clinic.
We welcome the opportunity to reply to the points raised by Hutchinson and his colleagues concerning our recent paper and to provide further scientific evidence that patients attending a multidisciplinary ALS clinic have improved survival compared with patients attending a general neurology clinic.
The key criticism is that the survival benefit derived from attending the ALS clinic is a result of referral centre bias. Hutchinson et al maintain that the multidisciplinary ALS clinic selects for patients with milder disease, as only these patients live long enough to be referred to the ALS clinic. While we acknowledge that it is challenging to avoid referral bias when one is quantifying the effect of a referral centre, referral bias is not a prominent factor in our study, for the following reasons. First, survival analysis of patients diagnosed exclusively in the ALS clinic (that is, not referred from other neurologists) reveals a similar beneficial effect on mortality compared with the previously published data (median survival, 644 and 448 days for the ALS clinic (n = 44) and general neurology cohorts, respectively; log-rank test, p = 0.02); Second, ALS patients who live more than a year from the time of diagnosis (and therefore have ample opportunity for referral) continue to experience a survival advantage from attending the ALS clinic.
Hutchinson et al suggest that that the parallel nature of the survival curves (that is, the similar rate of decline of both cohorts) stems from the overrepresentation in the general neurology clinic of more disabled patients who die in the first 250 days and “are not well enough to attend the ALS clinic.” Consequently, the perceived difference in mortality is artefactual. In reality, the parallel nature of the survival curves provides the strongest proof of a robust improvement in survival along ALS patients attending the ALS clinic: survival curves would converge rather than remain parallel, if the improved survival observed in the ALS cohort reflected the early loss of sicker patients in the general neurology clinic cohort. Kaplan–Meier survival analysis demands that the death of each individual be an independent event. Therefore, the prognosis of an individual who dies two years after diagnosis is unrelated to that of all other ALS patients who die within the first 250 days of diagnosis. The vast majority of positive intervention studies in both ALS patients and transgenic ALS mouse models show a parallel pattern of survival curves, as is seen in our paper. This common finding is thought to reflect the ability of an intervention to slow down, but not reverse, the progression of motor neurone degeneration that underlies ALS.
A single visit to the ALS clinic is associated with an improved survival of the ALS patient. The term “ALS clinic” is a misnomer. In fact, it represents a system of care that “services the Irish ALS population by combining the existing infrastructure of community services and the services of a voluntary organisation with a hospital based system.”1 The primary advantage of all “multidisciplinary clinics” is the coordination of a network of hospital and community based ancillary services (including respiratory medicine, nursing, occupational and physical therapy, speech and swallowing, nutrition, home help, counselling, and so on) that facilitate symptomatic interventions for each ALS patient, both in a hospital setting and in their home.2 Therefore, any patient who attends the clinic on a single occasion is enrolled in this system and is assiduously followed up. When a patient becomes too ill to travel to the clinic, home visits are undertaken by a specialist ALS nurse who coordinates and integrates community based, hospital based, and, in the latter stages, hospice based care. The improved survival observed in patients who attend the clinic on one occasion is thus a testimony to the “ALS clinic” system and contradicts the assertion that the observed difference in survival arises from the exclusion of ALS patients who are not fit enough to travel.
In our opinion, the criticism of the use of Kaplan–Meier survival curves rather than tables to present survival data is not valid. The vast majority of modern peer reviewed journals, including JNNP, do not publish survival tables, as the graphic representation of survival curves provides a greater wealth of data.
Similarly, if the baseline characteristics of patients attending the multidisciplinary clinic are solely responsible for our findings, attendance at the ALS clinic would not be independently predictive of survival in the Cox proportional hazards model. The Cox proportional hazards model is a popular mathematical model that allows estimation of hazard ratios and survival curves, even though the baseline hazard is not specified. Furthermore, it has been established that the site of onset, age, sex, and delay in diagnosis are surrogate markers of ALS disability.
The purpose of our study was to determine the optimum method of providing care to ALS patients. We agree that a randomly assigned study in which age, sex, site of onset, and disability are matched for each cohort would be ideal to demonstrate a difference between two different clinic types. However, this could only be accomplished in the setting of a formal randomised clinical trial, which would be both logistically difficult and ethically questionable.