Institution and research group
Authors: Tero Kivelä1, Sophie Piperno-Neumann2, Laurence Desjardins3, Alexander Schmittel4, Nikolaos Bechrakis5, Edoardo Midena6, Serge Leyvraz7, Leonidas Zografos8, Jean-Daniel Grange9, Guillaume Ract-Madoux9, Ernest Marshall10, Bertil Damato11 and Sebastian Eskelin1 for the European Ophthalmic Oncology Group
Affiliations: 1Ocular Oncology Service, Department of Ophthalmology, University of Helsinki and Helsinki University Hospital, Helsinki, Finland; 2Department of Medical Oncology and 3Department of Ocular Oncology, Institut Curie, Paris, France; 4Department of Hematology and Medical Oncology and 5Departments of Ophthalmology, Charité, Berlin, Germany and Innsbruck Medical University, Innsbruck, Austria; 6Department of Ophthalmology, University of Padova, Padova, Italy; 7Department of Oncology and 8Hopital Jules Gonin, University Hospital, Lausanne, Switzerland; 9Department of Ophthalmology, Croix-Rousse Hospital and Centre Léon Bérard, Lyon, France; 10Clatterbridge Centre for Oncology, Bebington, Wirral, Merseyside, United Kingdom; 11St Paul's Eye Unit, Royal Liverpool University Hospital, Liverpool, United Kingdom
Data set and variables
The model (currently called Helsinki University Hospital Working Formulation) was built on the basis of a multivariable analysis of comprehensive data on 54 patients from the then Helsinki University Central Hospital (HUCH) who died of metastatic malignant uveal melanoma between January 1985 and December 2000 (https://www.ncbi.nlm.nih.gov/pubmed/12518371). This national referral center manages about 90% of primary uveal melanomas in Finland. The model has thereafter been used as one prognostication tool in this institution.
The model was later externally validated (https://www.ncbi.nlm.nih.gov/pubmed/27296487) using data of 249 European patients from seven member institutions of the European Ophthalmic Oncology Group (OOG; http://www.oogeu.com). Most patients in the building and validation data sets participated in 6 to 12-monthly surveillance to detect metastases early that included liver imaging and liver function tests chosen according to the preference of each center. Most patients in both data sets also received some type of active therapy after detection of metastases.
The OOG has made the validation data open access through the Zenodo repository: https://zenodo.org/communities/oog/
The variables required for the model are 1) performance index as an estimate of the general health of the patient (such as Karnofsky index or the Eastern Cooperative Oncology Group [ECOG] score, also known as the WHO performance score); 2) the largest diameter of the largest metastasis as an estimate of measurable metastases (the best estimate from imaging studies such as ultrasonography, computed tomography or magnetic resonance imaging); and 3) the serum or plasma alkaline phosphatase level as an estimate of hepatic function and unmeasurable metastases (relative to its corresponding upper normal limit).
The model assigns each patient in one of three prognostic groups based on predicted overall survival, calculated on the basis of the covariates of the patient and the baseline survivor function of the original Cox multivariable regression: stage IVa (predicted median survival ≥12 months), stage IVb (predicted median survival <12-6 months) and stage IVc (predicted median survival <6 months).
In the validation dataset of 249 patients, the observed median survival was 18.6 months (95% CI 16.3-21.1) for stage IVa, 10.7 months (95% CI 8.6-14.0) for stage IVb, and 4.6 months (95% CI 3.0-6.7) for stage IVc. Of stage IVa patients, 74% survived for ≥12 months. In stage IVb, 76% of patients survived for ≥6 months and 44% survived for ≥12 months. In stage IVc, 63% of patients died in <6 months and 90% in <12 months.
The model has the following limitations. 1) It only predicts survival of patients with newly diagnosed metastases, and it cannot be applied to patients about to undergo second or higher line treatment. 2) The model does not accurately predict observed survival of patients who undergo resection of metastases. It is presumed that the largest metastases, which typically are resected when surgical treatment is selected, strongly influence survival. However, the model will suggest what would be the predicted survival of these patients in case that their metastases would be managed without resection. 3) The model is validated to categorize a cohort of patients by their predicted overall survival. It does not predict the observed survival category of a single individual. The model is best used for evidence-based, stage-specific analysis and reporting of overall survival of cohorts of patients with metastatic uveal melanoma in both prospective and retrospective treatment trials, and for stratification purposes in such trials.
References:
- Eskelin S, Pyrhönen S, Hahka-Kemppinen M, Tuomaala S, Kivelä T. A prognostic model and staging for metastatic uveal melanoma. Cancer 2003 Jan 15;97:465-75.
- Kivelä T, Piperno-Neumann S, Desjardins L, Schmittel A, Bechrakis N, Midena E, Leyvraz S, Zografos L, Grange J-D, Ract-Madoux G, Marshall E, Damato B, Eskelin S. Validation of a prognostic staging for metastatic uveal melanoma: a collaborative study of the European Ophthalmic Oncology Group. Am J Ophthalmol, 2016 Aug;168:217-226.
Statistics and system tools
Overall survival after detection of the first metastases from uveal melanoma was calculated from the date of diagnosis to death from metastases. Deaths due to other causes were not observed. A Cox proportional hazards model obtained from the building dataset was fitted (Table 1). With the β-coefficients of this model, a prognostic index (PI) is estimated. The baseline survival curve corresponding to the average PI value is computed, which is similar to the average survival for the total group. If the average PI is taken as baseline reference, the relative risk (RR) of an individual patient is given by RR = eestimated PI/eaverage PI. With the RR of an individual patient, an expected survival curve for that patient can be computed from the average survival curve corresponding to the average PI. If the median survival time of this curve is ≥12 months, the patient is assigned to stage IVa, if it is <12-6 months the patient is assigned to stage IVb, and if it is <6 months, stage IVc is assigned.
Table 1. Cox Multivariate Proportional Hazards Regression of Survival in Stage IV Uveal Melanoma
| Characteristic | Regression coefficient (standard error) |
Wald χ² | P | Hazard ratio (95% CI) |
|---|---|---|---|---|
| Performance indexa | 0.8069195 (0.3252891) | 6.15 | 0.013 | 2.24 (1.18–4.24) |
| Largest diameter of largest metastasisb | 0.1967883 (0.0660405) | 8.88 | 0.003 | 1.22 (1.07–1.39) |
| S-APc | 0.6092652 (0.2992219) | 4.16 | 0.042 | 1.84 (1.02–3.31) |
| Expected time on chemotherapyd | −0.1175717 (0.0312728) | 14.14 | < 0.001 | 0.89 (0.84–0.95) |
a Coding: Karnofsky index 100-90 or Eastern Cooperative Oncology Group Performance Status (ECOG) 0 = 1; Karnofsky 85-60 or ECOG 1–2 = 2; Karnofsky 50-0 or ECOG 3–4 = 3
b Continuous variable, in centimeters.
c Continuous variable, serum (or plasma) alkaline phosphatase (AP) divided by its upper normal limit
d Constant term, per month. It was not possible to determine whether more rapid than average progress of metastases is due to ineffective therapy or to more than average aggressiveness of the disease in the building dataset. Not only are patients with smaller metastatic lesions preferentially offered chemotherapy, they also may have less aggressive disease. If so, they are less likely to progress rapidly and more likely to stay longer on chemotherapy and also to receive second line treatments. Distinguishing between these two possibilities was not mandatory when time on chemotherapy was modeled as a confounding factor. Whether a long time on therapy was due to efficacy of treatment or to less aggressive disease, time on chemotherapy adjusts for the confounding and helps to quantitate contribution to prognosis of other modeled factors. Because the eventual time of chemotherapy is unknown for a new patient, the population mean of 5 months observed in the building data set is used in all new estimations.
Disclosure and sources of funding
The funding source had no role in the study design, collection, analysis, interpretation of data, or writing of the report. The project has been funded exclusively by academic, non-profit research funds. It was supported by the Helsinki University Hospital Research Fund (TYH2010204, TYH2012106, and TYH2013316) and the Sigrid Jusélius Foundation, Helsinki, Finland.