In patients with advanced lung cancer, regardless of the duration of their clinical experience, physicians overestimated the median survival time at diagnosis by almost twice the actual survival later observed (observed to expected ratio, 1.8 to 2.2). Younger physicians, registrars, and, to an even greater extent, residents were more accurate. Interestingly, a relationship between departure of prediction from actual survival and patient quality of life (global status and role functioning QLQ-C30 scales) was observed in younger physicians, suggesting a better knowledge or higher perception of their patients' actual prognosis, compared with senior doctors. Classical prognostic factors, such as a poor WHO PS score, low hemoglobin level, low protein level, high platelet count, and high LDH level, were confirmed to be related to shorter survival, suggesting no particular selection bias in this sample.
The large majority of already published data on this issue relies on the first approach, and deals with chronic diseases and cancers. In general, and for quite obvious reasons, the accuracy of prediction for patient (cancer or other illnesses) survival is much better when the estimate is made near death, whereas prognostication over a longer term seems more uncertain, as confirmed by Brandt et al. [22
], with a cohort of 515 terminally ill patients. In this respect, Christakis et al. [23
] observed that physicians overestimated survival in terminal cancer or HIV patients (in 63% of cases, doctors overestimated the survival with a mean predicted-to-observed survival ratio of 5.3). However, Chow et al. [24
] showed, in their literature review based on 12 articles, that clinical predictions tended to be inaccurate in the optimistic direction.
With respect to determinants of lung cancer patient survival, most papers, if not all, have evaluated the relationships among several clinical and biological factors, such as TNM staging, WHO PS score, symptoms (anorexia, dyspnea, loss of weight), platelet count, LDH level, hemoglobin level, and protein level, blood group, and oncogene expression by tumor cells, etc. This was shown, for example, by Moldway et al. [25
] in adenocarcinoma patients, for whom positive Bcl-2 staining and A+B+H antigen tumor staining was associated with longer survival. A review of the literature by Chow et al. [24
], based on 19 articles, on prognostic factors confirmed that WHO PS score is a prognostic factor for survival. With respect to quality of life, Herndon et al. [15
], using univariate analysis, confirmed the usual prognostic factors related to poor survival, but also observed that lower quality of life on QLQ-C30 subscales related to pain, appetite loss, fatigue, lung carcinoma symptoms, overall quality of life, and physical functioning scale was related to shorter survival.
Only a few papers have dealt with the actual evaluation by physicians of each patient's likely survival. In this respect, Viganó et al. [26
], in a study of two cohorts of terminal cancer patients, suggested that physicians overestimated patient survival. In the first cohort, dyspnea, nausea, vomiting, liver metastases, and lung cancer were prognostic factors for survival. In the second cohort, the patients had worse outcomes than in the first cohort, and dyspnea, weakness, and breast, gastrointestinal, or urinary cancer were also prognostic for survival. The authors concluded that clinicians should focus on physical quality-of-life indicators to gather prognostic clues in these patients. Indeed, Coates et al. [28
], in a trial of breast cancer patients, found that there was a strong prognostic significance of quality-of-life scores after disease relapse in patients with advanced breast cancer. Gripp et al. [29
] carried out a study on 216 patients with terminal cancer and observed that PS, primary cancer, fatigue, dyspnea, use of strong analgesics, brain metastasis, leukocytosis, LDH level, blood level, and anxiety were related to survival, and that physicians' survival estimates were unreliable, especially in patients near death, a finding that diverges from findings observed in most published papers, as cited above. Moreover, they found that a strong doctor–patient relationship did not appear to improve the accuracy of the clinical prediction of survival. Llobera et al. [30
] saw, in 200 terminal cancer patients, that a shorter survival time was related to anorexia, weight loss, and dyspnea.
Obviously, our study has some limitations. First, the study population was drawn from patients admitted to only one chest department specialized in the management of lung cancer, and this reflects referral to a specialized setting. The number of patients is limited, but the study was conducted with well-defined types of patients with lung cancer. Moreover, the fact that classical prognostic factors were observed in this work suggests that our patient sample was representative, and that the exclusion of some patients for technical reasons was sound. Second, the restriction to advanced stage lung cancer with a poor prognosis was the consequence of the question asked, and of the knowledge at the start of the study that evaluation of patients with shorter survival times is more accurate. Finally, quality of life during admission in these patients might have been influenced by the acute, severe, but mostly reversible, conditions that had led to the hospital admission, rather than the baseline condition solely related to cancer. The former can be a confounding factor and needs to be explored further.
The observation of a difference in accuracy of survival estimates among the three categories of physicians needs to be explained. Though all physicians overestimated patient survival, residents were more accurate. Only a few clinical trials have evaluated the prognostic value of quality-of-life scores. For example, Viganó et al. [27
] found that clinicians overestimated patient survival. Gripp et al. [29
] observed results similar to our findings. In their study, doctors in training and experienced doctors estimated patient survival, and a final estimation was decided by a consensus vote. In our study, there was a trend toward better accuracy for doctors in training than for experienced doctors. The residents seemed to take into account the patients' quality of life in their estimation of patient survival; this is likely to be a result, at least in part, of their closer relationship with patients, involving empathy. Residents are directly involved in the care of patients with lung cancer and have a better perception of their health than senior physicians. Another explanation, perhaps, is the fact that physicians have only a limited knowledge about the social or cognitive impact on health status.
Without a description and evaluation of more efficient means to prospectively evaluate each patient's survival, such prediction inaccuracy will remain a problem, and clinical trials will continue to include patients who are not correctly satisfying the expected conditions. Therefore, the full benefit of experimental treatment will not be assessed correctly.