Prognosis in amyotrophic lateral sclerosis

Methods.

For 4 years beginning in April 1990, we identified all incident cases of ALS occurring in residents of King, Pierce, and Snohomish counties in western Washington.40 These counties contain approximately 2.5 million people in 5,900 square miles. Patients were identified through a surveillance system that relied most heavily on neurologists and the Muscular Dystrophy Association. Eligible patients were 18 years or older and diagnosed with ALS by a neurologist because they had progressive motor neuron disease that affected both upper and lower motor neurons or only lower motor neurons. Patients were excluded if they had primary lateral sclerosis or if a diagnosis other than ALS was made during follow-up.

All 180 patients identified were invited through their treating physician to participate in a case-control study that involved an in-person interview. A total of 174 patients agreed (97%). During the interview, which was conducted an average of 5.4 months after diagnosis (median 4 months), the patient provided information about demographic factors, cigarette smoking, lifetime physical activity, onset and severity of disease, and general health status. Income was missing for nine patients and was imputed, using predicted values from a linear regression model with age, sex, education, and occupation as independent variables. Body mass index (weight/height² in kg/m²) was calculated from patient report of height and weight at various times relative to the date of diagnosis. Site of disease onset was classified as extremity only or any bulbar involvement. Functional impairment due to ALS was graded with two simple measures at the time of the interview. The ALS Severity Score provided an ordinal score that ranged from 0 (worst) to 40 (best) based on the patient’s self-reported performance in the four categories of speech, swallowing, and lower and upper extremity function.46 The second measure used phonation time as a reflection of vital capacity.47 On a single breath, patients were instructed to say “ah” for as long as possible. This test was repeated three times, and the longest of the three phonation trials was used for analysis. General health status was assessed at the time of the interview with a standard questionnaire, the Medical Outcomes Study Short Form 36 (SF-36).48,49⇓ The SF-36 consists of 36 questions. Results are presented in eight dimensions with scores that range from 0 for poor health to 100 for good health, and two summary scores of mental and physical health.50 Scores for ALS severity, phonation time, and general health were made dichotomous using the median score to compare patients who had a high score (above the median) with those who had a low score (at or below the median). In situations where patients were unable to provide information directly (n = 20), surrogates provided as much information as possible, except for ALS Severity Score, phonation time, and the SF-36.

Patients who participated in the interview were telephoned every 3 months to ascertain medical follow-up and vital status. Upon verification of a patient’s death, all medical records were requested, including those of the six patients who declined to participate in the case-control study. Vital status was known as of December 1999 for all 180 patients. Based on the follow-up calls, all 180 patients remained eligible for the study, as defined above.

For analysis, time was measured in months from date of diagnosis to death from any cause, to tracheostomy, or to last follow-up. The effect of individual prognostic factors was assessed with Kaplan-Meier survival curves. Log-rank χ² statistics were estimated to assess equality of survival functions. Multivariable Cox proportional hazards regression analysis was performed to estimate adjusted hazard ratios (HR) and 95% CI associated with different prognostic factors. When the hazard for an individual factor was not proportional the multivariable models included an interaction term for time.51 For stepwise models, we used a p-to-enter of 0.05 and a p-to-remove of 0.10. All of these analyses were performed in SPSS for Windows (version 7.5, Chicago, IL).52 In addition, a classification and regression tree (CART) analysis was performed to identify those factors that most strongly predicted survival after diagnosis.53-55⇓⇓ CART analysis evaluates all potential predictors to find the single variable that divides the entire cohort into two groups that differ the most on a particular outcome, in this case survival. If a variable is continuous, such as age, all potential cut points are examined. The process is repeated on the resulting groups, again using all of the potential predictors. These analyses were performed in CART (version 2.0, San Diego, CA).

The Human Subjects Review Committee at the University of Washington approved the study.

Results.

As of December 1999, 168 of the 180 patients (93%) had died. The mean age at diagnosis was 61.3 years (median 64 years, range 25 to 87). Over the course of the study, only one patient received riluzole, and six became ventilator dependent. None of the 12 surviving patients was dependent upon a ventilator for survival. Information on use of gastrostomy tubes and noninvasive ventilation and on participation in treatment trials was not tabulated in this study. Median survival for all 180 patients was 32 months from onset of symptoms and 19 months from diagnosis. The 5-year survival after diagnosis was 7% (n = 13). Men survived longer than women, both from onset and from diagnosis (figure 1, A and B). The median survival from onset was 36 months for men and 29 months for women; from diagnosis, 23 months for men and 15 months for women. The Food and Drug Administration approved riluzole in December 1995, at which time 26 patients remained alive.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Cumulative survival for men and women in months (A) from symptom onset to death or (B) from diagnosis to death in 180 patients with ALS. Patients were censored at date of last follow-up. Six patients were considered to have reached the outcome of interest when they became ventilator dependent.

Table 1 provides the distributions for individual potential prognostic factors, with median survival in months and the proportion of patients surviving to 12, 24, and 36 months after diagnosis. An expanded version of table 1 is available at the Neurology web site (see table E-1 at www.neurology.org). As shown by the log-rank χ² statistics, survival differed significantly across categories of many factors. Given the importance of older age and female sex, HR and 95% CI were adjusted for age and sex. Any bulbar features at onset, shorter time from symptom onset to diagnosis, and ALS Severity Score were all associated with a significantly worse prognosis. A particularly powerful prognostic factor was the ALS Severity Score. For example, patients with scores at baseline greater than the median score of 31 had a median survival of 27 months compared to 14 months for those with lower scores. Other factors significantly associated with survival after adjustment for age and sex were county of residence and marital status. The trend was for longer survival among those with stable or increased body mass index from 5 years before diagnosis and among those with greater lifetime physical activity at work or during leisure time.

To assess the independent effects of well-established prognostic factors, we forced into a Cox proportional hazards regression model four key factors: age, sex, any bulbar features at onset, and time from symptom onset to diagnosis (Model 1 in table 2). All remained significantly and independently associated with survival. After adjusting for all other factors in the model, patients 65 years or older had over a threefold hazard compared to those younger than 45 years, and women had a 50% increased hazard compared to men. To identify the strongest independent prognostic factors, we formed Model 2 using forward stepwise selection of several potential prognostic factors that are listed in the footnote of table 2 and coded as in table 1. ALS Severity Score and phonation time were not included because they were missing for 23 patients. The estimated model showed that older age at diagnosis, any bulbar features at onset, nonwhite race, lack of a marital partner, and residence in King County were all independently associated with poorer survival. Sex and symptom onset to diagnosis, which were in Model 1, did not enter Model 2. Also of note, starting with all of the variables in Model 1 and allowing all of the other variables to compete in a stepwise fashion, only marital status and county of residence entered.

Because of the missing values, we considered ALS Severity Score and the phonation time in separate models. When added to Model 1 along with time from diagnosis to interview, ALS Severity Score >31—namely, higher functional level at the time of the interview—was significantly and independently related to survival (HR 0.4, 95% CI 0.2 to 0.5). Findings were similar for maximum phonation time greater than 13.3 seconds (HR 0.6, 95% CI 0.4 to 0.9). When both ALS Severity Score and phonation time were included in the same model, only ALS Severity Score >31 remained a significant prognostic factor for survival.

The CART analysis included all the factors considered in the stepwise Cox regression described above and in the footnote to table 2. This analysis identified age at diagnosis and time from symptom onset to diagnosis as the two most important predictors of survival in 174 patients who had information on all these variables. The most parsimonious tree that yielded the greatest level of prediction while minimizing misclassification had four terminal nodes, which list the median duration of survival from diagnosis and from onset of symptoms (figure 2). The first split occurred at diagnosis age of ≤39 years, with a subsequent split at time from symptom onset to diagnosis of ≤9.5 months, and a final split again on marital status. Patients who were 39 years or younger at the time of diagnosis (92% men) had the longest survival after diagnosis (median 59 months). Unmarried older patients with a longer time from symptom onset to diagnosis (70% women) had the shortest survival after diagnosis (median 9 months).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Classification and regression tree (CART) analysis for survival from ALS. Analysis based on survival from diagnosis although median survival from both diagnosis and symptom onset provided in each terminal node. Factors eligible for this model are the same as those eligible for Model 2 in table 2 (see footnote). The analysis identified the optimal breakpoint for continuous variables.

Results for the SF-36 are listed in table 3. Cox proportional hazards regression models were run separately for the eight dimensions and two summary scores, after adjustment for age, sex, months from symptom onset to interview, and ALS Severity Score at interview. For each dimension, except bodily pain, good health (high scores) was associated with longer survival than poor health (low scores) with adjusted HR less than 1.0. Adjusted HR were significantly less than 1.0 for general health, physical function, vitality, role emotional, and social function. The dimension with the strongest association with survival was physical function, followed by social function. When all eight dimensions were forced into the same model, only physical function remained independently associated with survival even after adjustment for the other prognostic factors indicated above. Neither of the summary measures was significantly related to survival when considered individually, but when included in the same model, good health on physical health summary was significantly associated with longer survival than poor health (adjusted HR 0.50, 95% CI 0.26 to 0.98). A similar trend was seen for mental health but was not significant (adjusted HR 0.75, 95% CI 0.48 to 1.17).

Discussion.

In this population-based cohort of 180 patients with ALS, we found a median survival of 32 months from onset of symptoms and 19 months from diagnosis. The 5-year survival after diagnosis was 7%. In previous population-based studies, the median survival from onset ranged from 23 months to 36 months, and from diagnosis, from 12 months to 23 months.1-19⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓ In these studies, 5-year survival ranged from less than 5% to 30%, the broad range perhaps reflecting differences in the methods used to identify patients with ALS, differences in criteria for diagnosis, and the small numbers of patients surviving beyond the first few years yielding unstable estimates. This study is consistent with other studies done in the 1990s, which suggest that the current management of patients with ALS has not improved survival.18,19⇓

Most studies, including this one, have found that older age and any bulbar features at onset are independently associated with poorer survival. Like others, we also found a shorter time to diagnosis associated with a shorter survival, perhaps reflecting more aggressive and thus easily diagnosed disease.15,17,18,28,34,39⇓⇓⇓⇓⇓ Even after adjustment for these factors, survival was significantly shorter for women than men. This finding is consistent with the results from another study of incident patients,17 although some studies have failed to find such an association.12,14,15,18⇓⇓⇓

Much of the information that we collected from patients as part of a case-control study seeking etiologic risk factors for ALS has not been examined previously for prognostic importance. Most other population-based studies have relied exclusively on retrospective review of medical records. When considered individually without adjustments, several other factors were associated with survival. After controlling for age and sex, only lack of a marital partner, residence in King County at time of diagnosis, and a measure of ALS severity remained associated with a poorer prognosis. Although the trend was for those with greater lifetime physical activity to have longer survival, associations were not significant after controlling for age and sex. We assessed the severity of the ALS at the time of the interview using a simple measure of ALS severity46 and of pulmonary function.47 As in other studies of such measures, they were associated with survival, even after adjusting for other established prognostic factors and time from diagnosis to interview.32,39,56⇓⇓

We used two different statistical techniques to identify which among the several potential prognostic factors were significantly and independently related to survival. Using Cox proportional hazards regression, we found in a stepwise model the independent predictors of shorter survival to be older age, any bulbar features at onset, nonwhite race, lack of a martial partner, and residence in King County. We do not know why residence in King County was significantly associated with poorer survival. The other technique, CART analysis, uses a different approach to identify the most important predictors. This analysis identified age, time from symptom onset to diagnosis, and marital status as the key prognostic variables. The few patients (n = 13) who developed ALS at 39 years or younger were 92% men and had the longest survival from diagnosis (median 59 months). Conversely, unmarried older patients with a longer time from symptom onset to diagnosis (70% women) had the shortest survival after diagnosis (median 9 months). In part, the poor prognosis for these patients may be related to a delay in diagnosis because their median survival from symptom onset was 34 months. Perhaps a spouse’s recognition of the problem and insistence upon an evaluation leads to an earlier diagnosis. Of note, regardless of whether Cox regression or CART were used, sex was not included in these models, suggesting that the effects of sex on survival are likely mediated through these other factors. Results of these multivariable models need validation in other databases before being considered in individual patients.

Psychological well-being has been associated with survival.20,21⇓ We evaluated patients with a general health status measure, the SF-36.48,49⇓ It has been used previously in patients with ALS57-61⇓⇓⇓⇓ and found to be a reliable and responsive measure.58 The dimension with the strongest association with survival was physical function. Of note, this dimension has been shown to correlate with lower extremity force megascores on the Tufts Quantitative Neuromuscular Exam.58 We were unable to confirm with the SF-36 the finding of the previous studies concerning psychological well-being. When both SF-36 summary measures were forced into the same model, we found that the physical health summary was significantly and independently associated with survival, whereas only a trend was present for the mental health summary. Possibly other disease-specific measures of psychological and emotional well-being would be associated with survival in patients with ALS.

These results reinforce the findings of other population-based studies in identifying older age, female sex, any bulbar features at onset, time from symptom onset to diagnosis, disease severity, and marital status as important prognostic factors in ALS. These factors would be essential to consider in evaluation of a new treatment because they could have a larger effect on outcome than the treatment and, if imbalanced among treatment groups, could confound the results of a clinical trial. Disease-specific measures of health status and objective measures of muscle strength would likely improve predictions of survival. This study concentrated on information available around the time of the diagnosis. Measures taken repeatedly during the early course of disease and characterization of the type of disease—for instance, primarily a lower motor neuron syndrome—would likely yield improved estimates of survival.