- © 2011 by American Society of Clinical Oncology
Clinical Dutch-English Lambert-Eaton Myasthenic Syndrome (LEMS) Tumor Association Prediction Score Accurately Predicts Small-Cell Lung Cancer in the LEMS
- Maarten J. Titulaer,
- Paul Maddison,
- Jacob K. Sont,
- Paul W. Wirtz,
- David Hilton-Jones,
- Rinse Klooster,
- Nick Willcox,
- Marko Potman,
- Peter A.E. Sillevis Smitt,
- Jan B.M. Kuks,
- Bart O. Roep,
- Angela Vincent,
- Silvère M. van der Maarel,
- J. Gert van Dijk,
- Bethan Lang and
- Jan J.G.M. Verschuuren
- From the Leiden University Medical Center, Leiden; Haga Hospital, the Hague; Erasmus University Medical Center, Rotterdam; University Medical Center Groningen, Groningen, the Netherlands; John Radcliffe Hospital, Oxford; and Queens Medical Centre, Nottingham, United Kingdom.
- Corresponding author: Maarten J. Titulaer, MD, PhD, J3-R-166, Leiden University Medical Center, PO Box 9600, 2300 RC Leiden, the Netherlands; e-mail: m.j.titulaer{at}lumc.nl.
-
Presented in part at the 20th Meeting of the European Neurological Society, Berlin, Germany, June 19-23, 2010, and 12th International Conference of Neuromuscular Disorders, Naples, Italy, July 17-22, 2010.
Abstract
Purpose Approximately one half of patients with Lambert-Eaton myasthenic syndrome (LEMS) have small-cell lung carcinomas (SCLC), aggressive tumors with poor prognosis. In view of its profound impact on therapy and survival, we developed and validated a score to identify the presence of SCLC early in the course of LEMS.
Patients and Methods We derived a prediction score for SCLC in LEMS in a nationwide cohort of 107 Dutch patients, and validated it in a similar cohort of 112 British patients. A Dutch-English LEMS Tumor Association Prediction (DELTA-P) score was developed based on multivariate logistic regression.
Results Age at onset, smoking behavior, weight loss, Karnofsky performance status, bulbar involvement, male sexual impotence, and the presence of Sry-like high-mobility group box protein 1 serum antibodies were independent predictors for SCLC in LEMS. A DELTA-P score was derived allocating 1 point for the presence of each of the following items at or within 3 months from onset: age at onset ≥ 50 years, smoking at diagnosis, weight loss ≥ 5%, bulbar involvement, erectile dysfunction, and Karnofsky performance status lower than 70. The area under the curve of the receiver operating curve was 94.4% in the derivation cohort and 94.6% in the validation set. A DELTA-P score of 0 or 1 corresponded to a 0% to 2.6% chance of SCLC, whereas scores of 4, 5, and 6 corresponded to chances of SCLC of 93.5%, 96.6%, and 100%, respectively.
Conclusion The simple clinical DELTA-P score discriminated patients with LEMS with and without SCLC with high accuracy early in the course of LEMS.
INTRODUCTION
Lambert-Eaton myasthenic syndrome (LEMS) is an autoimmune disorder affecting the neuromuscular junction, characterized by proximal muscle weakness, loss of tendon reflexes, and autonomic dysfunction.1 Antibodies are directed against P/Q-type voltage-gated calcium channels (VGCC).2 In 1956, Lambert et al3 first described this disease in association with lung cancer. Small-cell lung carcinomas (SCLC) are found in more than half of patients with LEMS.1,4–8 In those with SCLC (SCLC-LEMS), LEMS is initiated by an immune reaction to VGCC expressed on the surface of the SCLC.2 The resulting antibodies react with VGCC at the presynaptic nerve terminal of the neuromuscular junction causing neurologic dysfunction. The remaining patients with LEMS have no evidence of SCLC even after many years' follow-up and cancer screening. These patients are thought to suffer from a nonparaneoplastic autoimmune LEMS (nontumor LEMS [NT-LEMS]).
An SCLC has a profound impact on therapy and prognosis. Therapy is focused on the tumor, and survival of SCLC is poor, whereas patients with NT-LEMS are thought to have a normal life-expectancy. A diagnosis of LEMS prompts a careful search for an SCLC. Screening for SCLC by computed tomography (CT) –thorax, [18F]fluorodeoxyglucose (FDG) positron emission tomography (PET), and bronchoscopy is required, and as SCLC may be too small to detect initially, these investigations need to be repeated every 6 months for at least 2 years.9 With this approach, 96% of SCLC are found within a year of diagnosing LEMS,9 while a delay of more than 2 years is extremely rare; nevertheless, some patients undergo many procedures with associated anxiety and costs. Care would be improved if those who require intensive investigations as well as those no longer at risk could be identified.
There are several factors that may help. Older age,1,10 smoking,1,11 development of multiple clinical symptoms within 6 months after onset,12,13 raised erythrocyte sedimentation rate (ESR)1 and Sry-like high-mobility group box protein 1 antibodies (SOX1),14,15 all suggest the presence of SCLC. Its absence is suggested by young age, no smoking history, slow evolution of clinical symptoms, and the 8.1 HLA haplotype.13,16–18 However, it is not clear whether these variables are independent, nor which are most reliable. At present, none seems robust enough on which to base far-reaching decisions. We developed a prediction rule based on a combination of these variables that would distinguish SCLC-LEMS and NT-LEMS early in the course of disease, with sufficient accuracy to guide clinical decisions. Two large cohorts of patients with LEMS were studied to first derive the score and secondly to validate it.
PATIENTS AND METHODS
Derivation Sample
We included all Dutch patients with LEMS with and without SCLC seen between 1990 and 2006. Nationwide referral to Leiden began in July 19988 and all patients alive at the time were investigated in a standardized fashion; new patients were added prospectively.
The diagnosis of LEMS was based on characteristic clinical features of proximal muscle weakness, reduced tendon reflexes, autonomic symptoms, and in addition, either the presence of VGCC antibodies or characteristic features of repetitive nerve stimulation.1,19 Repetitive nerve stimulation supported the diagnosis when it showed a low compound muscle action potential amplitude, combined with a decrement at low-rate stimulation and an increment higher than 100% after high-rate stimulation or maximal voluntary contraction.20 The study was approved by local ethics committees in the Netherlands and in the United Kingdom After obtaining informed consent, we (M.T., P.W., P.S.S., J.K., or J.V.) interviewed and examined all patients using a structured checklist, except for 13 whose data were obtained from hospital records and referring neurologists.
Parameters of the Derivation Sample
Symptoms present within the first 3 months from onset of disease were used for the purpose of this study, as our goal was to develop a score that is valuable early in the course of disease. Information on potential predictors were collected.
Demographic items, including age at onset, smoking history, and weight loss.
The cutoff for age at onset was 50 years, as SCLC-LEMS is more common than NT-LEMS after then.11 Patients were considered smokers if their lifetime consumption was more than 100 cigarettes.21 Weight loss was dichotomised at loss of 5% of body weight within the first 3 months, based on former research on SCLC patients.22
Karnofsky performance status.
Karnofsky performance status is an attempt to quantify cancer patients' general well-being and their quality of life.23 The score runs from 100 to 0, where 100 is perfect health and 0 is death. A cutoff value of 70 was chosen; those with lower values need at least occasional assistance for activities of daily living.
Clinical features.
Clinical features included the time of appearance, as described before.12 Disease onset was defined using the first neuromuscular symptom, usually proximal leg weakness. While autonomic symptoms or fatigue may occur before, these features were often incompletely described and their date of onset proved difficult or impossible to determine.
Definition of SCLC-LEMS or NT-LEMS
Labeling a patient as SCLC-LEMS required histologic or cytologic proof of SCLC, or for NT-LEMS, a follow-up of at least 3 years after diagnosis of LEMS. Patients without detectable tumors and follow-up shorter than 3 years were excluded. With current screening strategies, the chances of missing SCLC after 3 years was considered remote.12
Validation Sample
The score was validated in a cohort of British patients with LEMS from Oxford and Nottingham, seen between 1986 and 2009. All patients were seen by the late John Newsom-Davis, MD, PhD, FRS (Fellow of the Royal Society of London), or by one of the authors (P.M. or D.H.J.). Both centers have specific expertise in LEMS. A complete data set was compiled, as described above.
Statistical Analysis
In the derivation sample, univariate logistic regression was used to assess each variable's predictive value for SCLC. χ2 analysis was performed, if variables were uniformly present or absent. Factors significantly associated with SCLC in univariate analyses were further included in multivariate analyses.
We performed hierarchical multiple regression based on likelihood ratios. As many variable performed equally well as assessed by their likelihood ratios, the final selection of variables in the score was based on clinical and practical issues.
Where clinical and laboratory data were comparable, preference was given to clinical parameters. If two or more variables were equally associated with presence of SCLC, the factor most easily collected was retained. Imputation of missing values was performed, but since it did not have a significant effect, this was not used. A score was created based on the relative magnitude of the fitted coefficients in the linear predictor of the logistic regression model. This score was compared with a score in which all variables were assigned equal importance. If equally sensitive, preference was given to the unweighted score for sake of simplicity. How well the score distinguished between SCLC-LEMS and NT-LEMS was quantified with the area under the curve (AUC) of the receiver operating characteristic (ROC).25 For external validation, the model was applied to the validation data set. The validation data were not analyzed until the model had been created. Percentages of patients with SCLC, predicted by the prediction model, were calculated using combined datasets to assess reliability. A prior risk of SCLC in LEMS of 50% was used.1,4–8 Statistical analyses were done with SPSS for Windows 16 (SPSS Inc, Chicago, IL).
RESULTS
The derivation sample included 109 Dutch patients with LEMS. Two patients were excluded because follow-up was shorter than 3 years. Fifty-eight patients (54%) had an SCLC, and 49 had no tumor (Fig 1). Median follow-up of NT-LEMS patients was 7 years (range, 3 to 40 years). The following parameters from the patients' histories differed significantly in univariate analyses: age at onset, weight loss (≥ 5% within the first 3 months), Karnofsky performance status, and smoking-related factors (smoking ever, smoking at onset, and pack-years; P < .001), as well as sex (P = .021, Table 1). The clinical symptoms that were more frequent in patients with SCLC-LEMS within 3 months were (Table 1): weakness in distal leg muscles (P = .001), arms (proximal or distal; P < .001), bulbar muscles (dysarthria, swallowing, chewing, or neck weakness; P < .001), male sexual impotence (P = .007), dry mouth (P = .005), and micturition difficulties (P = .019). For laboratory values, raised ESR (P = .023), abnormal hemoglobin level (P = .018), leukocyte count (P = .004), and presence of SOX1 antibodies (P < .001) differed significantly between NT-LEMS and SCLC-LEMS (Table 1). HLA-DR3 was the only factor seen more frequently in NT-LEMS (P = .008).
CONSORT diagram summarizing all patients from the two cohorts. (A) Dutch patients; (B) British patients. LEMS, Lambert-Eaton myasthenic syndrome; SCLC, small-cell lung cancer; NT, nontumor.
Characteristics of the Derivation Set of 107 Patients With LEMS
Formation of the Score in the Derivation Sample
In the derivation set, 4% of data were missing. For multivariate analysis, data were available for 100 patients (52 with SCLC-LEMS). Age at onset, smoking at onset, weight loss, bulbar symptoms, male sexual impotence (within 3 months), and SOX1 antibodies proved the main predictors of SCLC in patients with LEMS (Table 2). As Karnofsky performance status lower than 70 was unequivocally associated with SCLC, we also added it to our score.
Multivariate ORs From Logistic Regression Models for the Prediction of SCLC
The AUC using all factors was 96.6%. Omitting weighting of factors had a negligible effect (AUC, 96.2%). Weighting, that reduces the easy applicability of the score, was therefore not used. To investigate the accuracy of a score restricted to clinical and epidemiologic parameters, the AUC was calculated without SOX1 antibodies. The resulting AUC of this 6-point score was slightly lower at 94.4%, but since this approach did not alter appreciably the proportions of patients with low-risk or high-risk scores, this purely clinical score was chosen to be used in the validation sample (Table 3).
Dutch-English LEMS Tumor Association Prediction Score
Evaluation of the Score in the Validation Sample
The validation sample consisted of 122 British patients of whom 112 patients were included, 13 from Nottingham and 99 from Oxford (Fig 1). Ten patients were excluded because they had short follow-up (n = 6) or had no or incomplete tumor screening (n = 3). One patient was excluded because of coexistence of myasthenia gravis, since precise onset of clinical features could not be determined. Of these 112 patients, 46 had an SCLC (41%; P = .05 v the Leiden cohort). Univariate analyses of the United Kingdom cohort and both cohorts combined can be found online (Appendix Tables A1 and A2, online only). The two cohorts were very similar. Some variables seemed different. SOX1 antibodies were found in 45% of British SCLC-LEMS patients versus 66.7% in the Dutch (P = .26) and male sexual impotence in 52.9% in the United Kingdom versus 72.4% in the Netherlands (P = .18).
All data were available for 100 British patients with LEMS, 35 with SCLC. Multivariate odds ratios proved comparable to those in the derivation set (Table 2). The discriminatory abilities of the score in the second cohort were very good as well, with an AUC of 94.6%.
We created the Dutch-English LEMS Tumor Association Prediction (DELTA-P) score, combining both cohorts (n = 200). Each item is scored equally, creating a score ranging from 0 to 6, directly correlating with increasing risk of SCLC (Table 3), as shown in Figure 2 (Appendix Table A3, online only). A score of 0 or 1 virtually excludes an SCLC with a risk of 0% or 2.6%. A score of 3 to 6 should alert the physician to screen vigorously, as the risk for SCLC rises from 83.9% to 93.5%, 96.6% and even 100%, respectively. To illustrate the use of our score, two patients with LEMS are discussed in the Appendix (online only).
Predicted fraction of small-cell lung cancer (SCLC) in patients with Lambert-Eaton myasthenic syndrome (LEMS) based on the Dutch-English LEMS Tumor Association Prediction (DELTA-P) score. Point sizes proportionate to the number of patients with a specific score, also represented by the percentage inside the circle. Vertical bars indicate SEM.
DISCUSSION
Prognosis and treatment of LEMS differs greatly depending on whether the patient has an SCLC or not. We have developed and validated a simple clinical scoring system, the DELTA-P score, that can be easily applied in the clinic to predict with higher than 94% reliability the likelihood of SCLC in patients with LEMS early in the course of their neurologic disease. Based on age at onset, smoking at onset, weight loss, Karnofsky performance status, bulbar symptoms, and male sexual impotence, all within 3 months from onset, our score indicates the presence of SCLC with very high accuracy. This score provides physicians with a tool to identify high-risk patients, to reassure patients with very low risk, and to guide the screening process and follow-up.
In previous studies, many factors were associated with SCLC in patients with LEMS. Univariate analyses corroborated their associations with SCLC, all confirming our clinical impression that patients with SCLC-LEMS are more severely disabled more early in the course of disease than those without SCLC. Age at onset, smoking, and early involvement of clinical features (bulbar weakness, weakness in distal legs, proximal and distal arms, and male sexual impotence) had already been identified as predictors for SCLC.1,10,12 Some patients in these studies were included in our derivation10,12 and validation sets.1 ESR, HLA-B8, and HLA-DR3 differed significantly in univariate analyses,1,10,17 but had no independent discriminatory value in multivariate analyses. The HLA-B8 and HLA-DR3 association is only evident in NT-LEMS presenting before 50 years (own data, not published), which explains why HLA does not add to our score. SOX1 antibodies14,15 remained an independent predictor in multivariate analysis, as the AUC was 1.8% higher when they were included. However, SOX1 antibody tests are not commercially available yet and rather time-consuming. Besides, their inclusion in the score did not affect the percentage of patients with scores of either 0 or 1 (very low risk) or 4 to 6 (very high risk). Differences were found for patients with a score of 3 (22% of patients), subdividing the group in two high-risk groups, which would have no impact on screening strategy. Therefore, we decided to omit SOX1 antibodies.
The use of easy-to-obtain directly available clinical characteristics, avoidance of weighting of factors, and the use of a double acronym (Table 3) will enhance the practical implementation of the DELTA-P score.
The Dutch and British cohorts were comparable for SCLC-LEMS as well as for NT-LEMS, with a few exceptions. British patients with NT-LEMS smoked less at onset, most likely because of the earlier decline in smoking habits in the United Kingdom than in the Netherlands, from 36.2% in the Netherlands versus 28% in the United Kingdom in 1992, and 29.1% versus 21% in 2007.26,27 Male sexual impotence was reported more often in the Netherlands than in the United Kingdom. Patients in the derivation sample were interviewed with a structured questionnaire, which might explain the higher frequency and less missing data of impotence and other signs of autonomic dysfunction. The prevalence of SOX1 antibodies was lower in the British cohort especially for sera older than 5 years, concerning patients who had been treated with chemotherapy or immunosuppressants. Sera from British patients with SCLC-LEMS, drawn within the past 5 years, were positive in 67%, comparable to published percentages.14,15
Overall, more data were missing in the SCLC-LEMS group because some patients had died before additional information could be gathered. However, these comprise a small fraction of our study group and as these were the more severely disabled patients, any resulting bias would almost certainly have led us to underestimate the potential of our score.
The association between LEMS and SCLC is clear, but reported percentages of SCLC-LEMS range from 40% to 70%. Such differences probably reflect patient ascertainment, whether by cross-sectional or prospective collection of patients, and selection bias. Neuromuscular centers see more patients with NT-LEMS, while neuro-oncology departments see more patients with SCLC-LEMS. In our cohorts, SCLC-LEMS was found in 54% and 41% in the derivation and validation sets, respectively. Whereas the Dutch cohort was collected nationwide, via both neuromuscular and neuro-oncology centers, nearly all United Kingdom patients were referred to the neuromuscular clinic in Oxford, explaining the lower percentages of SCLC-LEMS. Overall, 50% to 60% seems to be reasonable. For this, we used a prior chance of 50% for calculating the consequences of our DELTA-P score (Fig 2). This assumption hardly affects applicability of the score, as different prior chances (40% to 70%) only affected patients with a DELTA-P score of 2 (Appendix Fig A1, online only), who must be rescreened anyway and comprise only 15% of patients (Fig 2).
Although other tumors have been associated with LEMS,28 it is hard to distinguish between chance and cause. Among 113 patients with NT-LEMS included, only four tumors were found in three patients without any effect on their neuromuscular symptoms. One patient had breast cancer 6 years before onset of LEMS, followed by non-Hodgkin's lymphoma (NHL) 20 years after the diagnosis of LEMS. Endometrial carcinoma and NHL were found in a second and third patient, 7 and 12 years after onset of LEMS. Both instances of NHL were found in patients who had been treated with high-doses of immunosuppressants. According to recommendations of the Paraneoplastic Neurological Syndromes Euronetwork,29 these patients were regarded as patients with NT-LEMS. Their DELTA-P scores were 3, 0, and 1.
In 2008, we proposed a screening strategy based on data from our derivation set.9 Each patient should be screened by CT thorax and FDG-PET (or integrated FDG-PET/CT). If negative, this should be repeated after 6 months by CT thorax or FDG-PET, to be repeated every 6 months for 2 years. As overlooking a SCLC is not acceptable, we propose to screen every patient at least twice. The DELTA-P score can help to guide the need for further screening. In patients with a DELTA-P score of 0 or 1, we suggest to stop screening after two adequate and negative screens. The chance of developing a SCLC afterwards is lower than 1 per 1,000, even if calculated conservatively. If the score is 3 to 6, the second screening should be performed earlier, after 3 months, and should be repeated every 6 months afterward for 2 years. In fact, the physician should be continuously vigilant for signs pointing toward SCLC. In patients with a DELTA score of 2, screening should remain as proposed before, every 6 months for 2 years.9
In conclusion, SCLC can be predicted accurately in patients with LEMS, which will help to provide better clinical care to all patients. The DELTA-P score can be used in routine clinical practice to prioritize high-risk patients for intensive tumor screening, while it also identifies those with very low tumor risk, who can be reassured in an early phase of the screening process.
AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST
Although all authors completed the disclosure declaration, the following author(s) indicated a financial or other interest that is relevant to the subject matter under consideration in this article. Certain relationships marked with a “U” are those for which no compensation was received; those relationships marked with a “C” were compensated. For a detailed description of the disclosure categories, or for more information about ASCO's conflict of interest policy, please refer to the Author Disclosure Declaration and the Disclosures of Potential Conflicts of Interest section in Information for Contributors.
Employment or Leadership Position: None Consultant or Advisory Role: Jan J.G.M. Verschuuren, BioMarin (C) Stock Ownership: None Honoraria: None Research Funding: None Expert Testimony: None Other Remuneration: None
AUTHOR CONTRIBUTIONS
Conception and design: Maarten J. Titulaer, Paul W. Wirtz,Jan J.G.M. Verschuuren
Financial support: Maarten J. Titulaer, Silvère M. van der Maarel,Jan J.G.M. Verschuuren
Administrative support: Maarten J. Titulaer, Paul Maddison, David Hilton-Jones, Nick Willcox, J. Gert van Dijk, Bethan Lang,Jan J.G.M. Verschuuren
Provision of study materials or patients: Maarten J. Titulaer, Paul Maddison, Paul W. Wirtz, David Hilton-Jones, Rinse Klooster, Nick Willcox, Marko Potman, Peter A.E. Sillevis Smitt, Jan B.M. Kuks, Bart O. Roep, Angela Vincent, Silvère M. van der Maarel, J. Gert van Dijk, Bethan Lang, Jan J.G.M. Verschuuren
Collection and assembly of data: Maarten J. Titulaer, Paul Maddison, Paul W. Wirtz, David Hilton-Jones, Rinse Klooster, Marko Potman, Peter A.E. Sillevis Smitt, Jan B.M. Kuks, Silvère M. van der Maarel, Bethan Lang, Jan J.G.M. Verschuuren
Data analysis and interpretation: Maarten J. Titulaer, Jacob K. Sont,Jan J.G.M. Verschuuren
Manuscript writing: All authors
Final approval of manuscript: All authors
Acknowledgment
We dedicate this article in affectionate gratitude to the memory of John Newsom-Davis, MD, PhD, FRS (1932 to 2007), who contributed so much both to our understanding of the Lambert-Eaton myasthenic syndrome (LEMS) and to the United Kingdom cohort of patients.
We thank all neurologists who referred patients with LEMS to us.
Appendix
Patient 1 is a 62-year-old male patient, who was still smoking (40 pack-years). His Karnofsky performance status was 60 and he lost 8 kilograms (10%) within the first 3 months. He had a subacute onset of weakness in legs and arms, with rapid spreading to the feet and hands, though not to the bulbar region. He had autonomic symptoms from the start (impotence, micturition difficulties, and constipation). His DELTA-P score was 5, corresponding to an SCLC risk of 96%. He was screened by chest x-ray, CT-thorax, lumbar puncture, and FDG-PET scan, which were all reported to be normal. Three months later, magnetic resonance imaging of the thoracolumbar region, performed because of pain, showed a lesion, which was histologically proven to be an SCLC metastasis. Retrospectively, the spot was already visible on FDG-PET. He was treated with chemotherapy and radiotherapy, resulting in temporary remission. Twenty-eight months after onset, he developed a cerebellar metastasis. A primary lung tumor was never found. He was treated by resection and adjuvant whole-brain radiotherapy. He died 52 months after onset of symptoms due to his cerebral metastases. If the DELTA-P score and implied 96.6% risk of SCLC had been known, the initial screening would probably have been more rigorous.
Patient 2 was a 68-year-old woman who stopped smoking 8 years before the onset of symptoms (10 pack years). Karnofsky performance status was 90 and she had no weight loss. She presented with difficulties climbing stairs, dry eyes, and dry mouth, and did not develop bulbar symptoms for 7 months. She had a DELTA-P score of 1, corresponding to a predicted risk of SCLC of 2.6%. She died at the age of 92, without any sign of SCLC over 24 years of follow-up.
Characteristics of the Validation Set of 112 Patients With LEMS
Characteristics of 219 Patients With LEMS (both cohorts combined)
No. of Patients With LEMS With SCLC in the Derivation and Validation Sets According to the Dutch-English LEMS Tumor Association Prediction Score
Predicted fractions of small-cell lung cancer (SCLC) in patients with Lambert-Eaton myasthenic syndrome (LEMS), depending on different prior chances, based on the Dutch-English LEMS Tumor Association Prediction (DELTA-P) score.
Footnotes
-
Supported in part by a European Neurological Society fellowship, the Leiden University Medical Center, the European Union research project PNSEuronetwork, contract No. LSSM-CT 2005-518174, and the Prinses Beatrix Fonds.
-
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
- Received August 10, 2010.
- Accepted November 5, 2010.
