Prophylactic cranial irradiation in small cell lung cancer: a systematic review of the literature with meta-analysis

Purpose A systematic review of the literature was carried out to determine the role of prophylactic cranial irradiation (PCI) in small cell lung cancer (SCLC) . Methods To be eligible, full published trials needed to deal with SCLC and to have randomly assigned patients to receive PCI or not. Trials quality was assessed by two scores (Chalmers and ELCWP). Results Twelve randomised trials (1547 patients) were found to be eligible. Five evaluated the role of PCI in SCLC patients who had complete response (CR) after chemotherapy. Brain CT scan was done in the work-up in five studies and brain scintigraphy in six. Chalmers and ELCWP scores are well correlated (p < 0.001), with respective median scores of 32.6 and 38.8 %. This meta-analysis based on the available published data reveals a decrease of brain metastases incidence (hazard ratio (HR): 0.48; 95 % confidence interval (CI): 0.39 - 0.60) for all the studies and an improvement of survival (HR: 0.82; 95 % CI: 0.71 - 0.96) in patients in CR in favour of the PCI arm. Unfortunately, long-term neurotoxicity was not adequately described . Conclusions PCI decreases brain metastases incidence and improves survival in CR SCLC patients but these effects were obtained in patients who had no systematic neuropsychological and brain imagery assessments. The long-term toxicity has not been prospectively evaluated. If PCI can be recommended in patients with SCLC and CR documented by a work-up including brain CT scan, data are lacking to generalise its use to any CR situations.


Introduction
Small cell lung carcinoma (SCLC) has a very poor prognosis when untreated. The development of chemothera-py, with or without chest radiotherapy, has allowed to obtain survival improvement and a small percentage of cures. However the majority of the patients relapse and only <25 % of complete responders will be long-term survivors [1]. The central nervous system (CNS) is a frequent site of relapse. About 10 % of the patients initially present with brain metastases. The two-year cumulative risk rises to ≥ 50 % [2] and CNS metastases are found in up to 65 % of patients at autopsy [3]. The median survival time after brain metastases diagnosis is 4 to 5 months. Because the blood-brain barrier has been considered to protect the CNS from most cytotoxic agents and as SCLC is very radiosensitive, the role of prophylactic brain irradiation (PCI) has been studied in several trials. The results of the randomised trials show that PCI reduces the frequency of brain metastases although survival is not consistently improved. Some data suggest that the gain in survival is restricted to patients in complete remission (CR). A recently published meta-analysis [4] of PCI for SCLC in patients with CR after chemotherapy has analysed the data of 7 randomised studies (including one abstract and one unpublished study) concerning a total of 987 patients (526 treated with PCI and 461 controls). The relative risk (RR) of death in the treatment group as compared to the control group was 0.84 (95 % confidence interval CI: 0.73 to 0.97; p= 0.01). PCI decreased also the cumulative incidence of brain metastases (RR: 0.46; CI 95 %: 0.38 -0.57; p < 0.001). Unfortunately the authors have not mentioned the performance of cerebral imagery (CT scan or MRI) in the work-up or the follow-up and have not reviewed the cerebral toxicity of PCI. Some articles have dealt with this question. Johnson [5] reported 20 longterm SCLC survivors with a median follow-up of 6 years (2.4 to 10.6 y). Fifteen SCLC were treated by PCI, 2 by therapeutic cranial irradiation and 3 had no cranial irradiation. Fifteen had neurologic complaints (memory loss, walking or writing difficulties, weakness...), 15 had abnormal brain CT scan (ventricular dilatation, brain atrophy...) and 12 had abnormal mental status examination. Neurologic abnormalities seemed thus to be very common in long-term survivors SCLC and may be more prominent in patients having received high-doses chemotherapy or treated with large brain radiotherapy fractions. Lee [6] reported 3 cases of dementia, confusion and ataxia over 24 patients who received PCI. There was no toxicity in the control group. Toxicity appeared 2.5 years after PCI, the follow-up ranging from 37 to 74 months. In the Chake's study [7], five out of seven patients had progressive dysfunction leading to death in 1 to 26 months after PCI. Foncesca [8] related 14 % leucoencephalopathy in patients with SCLC who received PCI. The mean time of onset of symptoms was 357 days, the median follow-up time being 59 months. Symptoms consisted of intellectual change, memory loss and motor abnormalities. Laukkanen [9] related 60 % memory loss but no dementia two years after PCI. In the Licciardelo study [10], severe neurologic toxicities occurred in two of 15 patients (2.5 and 30 months after PCI). Finally, Van Oosterhout [11] reported no statistical evidence for additional neurotoxicity (follow-up of 2 years) in a series of 51 patients whatever they had received or not PCI. But there was difference in the neuro-psychological examination between patients and matched healthy controls, that might indicate that cognitive impairment is partly disease-related (probably due to emotional distress and deteriorated physical conditions). All these studies being taken in consideration, the problem of cerebral toxicity remains unclear, leading to controversy about the indications of PCI in SCLC.
The purpose of the present article is to assess the role of PCI in SCLC by performing a syste-matic review of the randomised trials published in the literature. A qualitative evaluation of their methodology was performed, including brain imagery work-ups and neuropsychological assessment as well as an aggregation (meta-analysis) of survival and brain relapse results.

Trials selection
To be eligible for the systematic review, trials needed to deal with SCLC exclusively, to have randomly assigned patients to receive prophylactic cranial irradiation or not and to have been published as a full paper in the French or English language literature before January 2000.
Articles were identified by an electronic search (Medline) using the keywords "small cell lung carcinoma" and "prophylactic cranial irradiation" completed by the personal bibliography of one of the authors and by the references reported in the selected studies.

Methodological assessment
To assess the trial methodology, nine investigators, including six physicians, one biostatistician, one biologist and one pathologist read each publication, guarantying the critical reading of the selected articles. They were then scored according to two quality scales: the score proposed by Chalmers et al. [12] and the score proposed by the European Lung Cancer Working Party (ELCWP) [13,14] as described in Appendix A. The participation of many readers was a guarantee for the correct reading of the articles. The Chalmers score evaluates two dimensions of quality: the internal (scientific) and external (generalisability of results) validities, with respectively maximal scores of 63 and 25 points (the total being 88 points). The ELCWP score assesses two quality aspects: the protocol design (as usually reported in the patients and methods section of the publication) and the analysis performance (as reported in the results section) with maximal scores of 70 and 80 points respectively (with an overall maximum of 150 points). Each item was quoted using an ordinal scale (possible values: 2, 1 or 0). When an item was not applicable in a trial, its theoretically attributable points were not taken into account in the total of the concerned category. As the items were defined by data that could objectively be found in the article and did not require a subjective judgement, the score of each item was consensually determined in meetings where at least two thirds of the investigators needed to be present. The final score was expressed in percentage ranging from 0 to 100 %, higher values reflecting a larger application of methodological standards.

Statistical methods
The results of a study were considered as "positive" if the p value for the statistical test comparing the survival distributions between arms was < 0.05 in favour of the experimental arm. In the other situations (statistically significant survival benefit for the control arm or non statistically significant difference in survival distributions), it was called "negative". The same method was used to evaluate the time to relapse in the brain. The correlation between the quality scores, or two other continuous variables, was measured by the Spearman ranks correlation coefficient. Its significance was assessed by testing a null hypothesis of equality to zero of this coefficient. Non-parametric Mann-Whitney (for binary varia-bles) or Kruskal-Wallis (for multiple classes variables) tests were performed to compare quality scores distributions according to the value of the considered discrete variable. For the quantitative aggregation of the survival results, we measured the treatment effect by the hazard ratio (HR) between the survival distribution, according to a method that we have previously reported [15]. For each trial, this HR was estimated by a method depending on the results provided in the publications. The most accurate method consisted to retrieve the HR estimate and its confidence interval from the reported results or to calculate them directly using parameters given by the authors: the confidence interval for the HR, the log-rank statistics or its p value or the O-E statistic (difference between numbers of observed and expected events). If not available, we looked for the total number of events and the log-rank statistic or its p value allowing calculation of an approximation of the HR estimate. Finally, if the exploitable data were in the format of graphical representations of the survival distributions, we extracted survivals rates at some specified times in order to reconstruct the HR estimate and its variance with the assumption that the rate of patients censored was constant during the study follow-up. By convention a HR < 1 implied a survival benefit for the experimental arm. The same method was used for time to relapse of the brain (assessing brain metastases incidence).  [16,17,19,23,27] evaluated the role of PCI in SCLC patients who had a complete response after induction chemotherapy. Five studies [18,21,24,25,26] assessed the role of PCI administered at induction chemotherapy in patients considered as free of brain metastases. In two studies [20,22], PCI was given as treatment consolidation at the end of chemotherapy before response evaluation. Seven trials [16,17,18,23,24,26,27] included SCLC patients at all stages of this disease and five studies [19,20,21,22,25] only limited disease. Brain CT scan was done at the initial staging work-up in five studies [16,17,20,23,27] and brain scintigraphy in six [18,21,22,24,25,26]. In one study [19], the staging was only based on clinical examination. There was also a significant difference for ELCWP score according to the year of publication (Rs = 0.71; p = 0.01), with better quality score for the new recent studies.
The most poorly described items of the ELCWP scale were the work-ups including neuropsychological tests (with a mean score of 22 %), the evaluation criteria (27 %) and the treatment description (33 %) for the internal validity, the prognostic factors for relapse (0 %) or for survival (0 %) and the description of the neurological toxicities (14 %) for the external validity.
Half of the individual studies reported an improvement of time to relapse in the brain assessing incidence brain metastases [16,18,19,25,26,27] in the PCI arm but none showed an advantage in term of survival. For the meta-analysis of brain metastases incidence, data were available in 10 trials. The hazard ratio (HR) was provided in 2, it was calculated from the logrank statistic and he number of events in 7 or from the brain metastases incidence curves in one. The meta-analysis revealed a significant decrease in the incidence of brain metastases when all the studies were considered (fig 1) with a hazard ratio (HR) of 0.48 (95 % CI: 0.39 -0.60) and when only pa-  4).
We also performed some subgroups analysis. The decrease in brain metastases incidence was also present in the subgroups of patients with initial PCI, limited disease, any stage disease or who had brain CT scan at initial staging and just before randomisation for PCI (table 3). Results were not significant for survival in patients with initial PCI, limited disease or who had no brain CT scan before randomisation. Statistical significance was marginal in patients with any stage disease or who had brain CT scan for initially staging or just before randomisation for PCI (table 4).
Toxicity was rarely adequately described. There was no data in four studies. In five trials, authors provided a short narrative description mentioning no or minimal toxicity (as alopecia...); Ohonoshi [23] reported one case of sevenlong-term disease-free survivors who had mem-

Figure 1
Results of the meta-analysis of the studies evaluating the role of PCI on time to relapse in the brain assessing brain metastases incidence : HR : 0.48 (95% CI : 0.39-0.60) NB: the centre of the lozenge gives the combined HR of the metaanalysis and its extremities the 95% confidence interval.

Figure 2
Results of the meta-analysis of the studies evaluating the role of PCI on time to relapse in the brain assessing brain metastases incidence when patients are in complete response : HR : 0.49 (95% CI : 0.39-0.62)

Figure 3
Results of the meta-analysis of the studies evaluating the role of PCI on survival : HR : 0.94 (95 % CI : 0.87-1.02)

Figure 4
Results of the meta-analysis of the studies evaluating the role of PCI on survival when patients are in CR : HR : 0.82 (95% CI : 0.71-0.96) ory disturbance and gait ataxia in the PCI arm. Two trials reported neuropsychological evaluation in a part of the randomised patients. Gregor [19] performed an assessment of cognitive function in 40 % of the randomised patients and showed no difference between the two arms at two years after PCI but without latter data. Arriagada [16] evaluated 60 % of the patients two years after PCI and showed no difference between the two groups.

Discussion
This systematic review, by pooling all randomised studies comparing treatment of SCLC with or without PCI, revealed a positive effect of PCI. As shown by the metaanalysis, PCI reduced brain metastases incidence and improved survival in patients in CR after chemotherapy (especially when brain CT scan was part of the staging work-up). Unfortunately, the performance of brain imagery (CT scan or MRI) and the long-term assessment of neuropsychological toxicities are not well described in the 12 available trials.
To perform a meta-analysis comparing such heterogeneous trials, we have used a methodology that was similar to our prior systematic reviews [14,15]. All trials were assessed by 9 investigators using two quality scores: the Chalmers and the ELCWP scales. The latter scale was adapted to the present topic by introducing some specific changes: in the work-up, brain CT scan or MRI with neuropsychological assessment was needed to have 2 points; in the treatment description, brain irradiation method had to be described; neuropsychological examination results were added in the patients characteristics and the "local control of tumour" item was changed in a "brain metastases incidence" item. The results obtained with the two scales were compared and a significant correlation was observed. There was no quality difference among the publications, allowing quantitative aggregation (meta-analysis) of the results of the individual trials.
The only significant finding in the performed comparisons was an improved quality in favour of more recently published trials, which can be explained by a better knowledge of clinical trials methodology standards over the last years.
Our approach does not however prevent all the potential biases. The most important one is probably the publication bias. Our review took into account only fully published studies. We did not look for unpublished trials and abstracts because the methodology used required data available in full publications only. Meta-analysis based on individual data is considered by some authors as the gold standard [28]. Systematic reviews of the literature and meta-analyses of individual patient data should not be confused. The first approach is only based on the fully published studies and provides an exhaustive and critical analysis of the topic with an adequate methodology based on the criteria of Mulrow [29] and with data aggregation (meta-analysis) when possible. The second approach is in fact a new study taking into account all performed trials on the topic, whatever published or not, requiring individual data update by the investigators. In that latter, publications are mainly used for identification purposes. Our meta-analysis, based on the published data, has allowed us to find the same results for patients in CR as Auperin et al [4] in their individual data meta-analysis. This point supports the validity of our approach. Another potential bias is the language problem: we have restricted our review to articles published in English or French. This selection could favour the positive studies that are most often published in English while the negative ones tend to be more reported in native language [30]. The method of extrapolation of HR needs also to be discussed. When HR were not reported by the authors, they were calculated from the data available in the article and, if not possible, they were extrapolated from the survival curves. This approach might have been associated with errors due to imprecision of the reading.
The brain work-up is often poorly documented. Only five studies reported brain CT scan in the initial evaluation and only in two of them, brain CT scan was done just before randomisation for PCI (when patients were in CR after chemotherapy). So, in the majority of the studies, the CR population could contain patients with asymptomatic brain metastases for which the delivered PCI was in fact a consolidation therapy. To be sure that there are no brain involvement, brain CT scan should have been done just before PCI. In addition, the CR status depends on the type of work-up performed and on the presence of lesions due to chest irradiation, explaining probably why some groups report small rates of complete response. Moreover, the recent development of MRI that could reveal smaller asymptomatic brain metastases will require an update of these trials in the next few years. Indeed, in contrast to prior literature which showed a prevalence of brain metastases at presentation of 10%, Hochstenbag et al found a prevalence of 24 %. This difference can been explained by the fact that the prevalence of 10 % is based on clinical signs and confirmation by brain imaging and, that in the Hochstenbag's study, MRI diagnosed 15 % brain metastases in neurologically asymptomatic patients [31].
The neuropsychological toxicity of PCI was only described in retrospective studies performed with a small number of patients. In our review, two randomised trials reported neuropsychological assessments that was performed only in a part of the patients and during the first years following PCI. They provided no data about longterm toxicity. It should be noted that other factors than radiotherapy toxicity can also contribute to neurological complications. Indeed old age, alcohol, anticancer drugs (vincristine, etoposide,...), paraneoplastic encephalomyelitis [32;33] and tobacco long-term use or can produce demential syndromes. Concomitant administration of some types of chemotherapy is considered to contribute to brain radiotherapy toxicity. The fractionation and the total dose of radiotherapy delivered to the brain can also influence the toxicity. Neurological toxicity may be reduced by using 2 Gy fractions (20 to 40 Gy) and by giving PCI after chemotherapy. All these factors were not analysed in our systematic review because of a total lack of data in the report of the results of the individual randomised trials.
In conclusion, the present systematic review indicates that PCI decreases brain metastases incidence and that PCI improves survival in SCLC patients in CR after chemotherapy. These effects were obtained in patients who had no systematic neuropsychological brain imagery assessments. The long-term toxicity has so far not been prospectively evaluated. If PCI can be recommended in patients with SCLC and CR documented by a workup including brain CT scan, data are lacking to generalise its use to any CR situations as some would like [34]. Particularly the potential benefits of PCI have to be carefully balanced with the possible long-term effects, in patients who are managed with more modern imagery techniques like MRI. New trials, adapted to these new developments, are necessary. -initial : brain CT scan or MRI and/or neuropsychological assessment -at response assessment (idem)