FROM ART TO SCIENCE
S. GARATTINI
Istituto di Ricerche Farmacologiche "Mario Negri" Via Eritrea, 62 20157 MILAN, Italy
The rational development of a drug for the treatment of lung cancer implies a certain amount of preclinical knowledge. Over the last decades considerable progress have been made and as a result, today, we have a large battery of tests for screening new drugs available.
In vitro. Several strains of lung cancer cells are available for growth in different media. They represent different histological types of human tumors and can be modified by transfection or knock-out techniques. These cells permit large scale screening. The NCI screening program has tested more than 100,000 chemicals; but the yield of substances selectively acting on lung cancer cells is still low.
In vivo. Most tests are carried out in mice, though sometimes in rats. There are several animal models, with different meanings. There are techniques for obtaining intrapulmonary tumor nodules; others for metastatic pulmonary nodules distal from the site of implantation of the primary tumor. Several chemicals can be employed to induce lung cancer in mice. These models are more stringent in mice with no immunological reactions (SCID or "nude" mice).
Recently lung tumors have been obtained in transgenic or knock-out mice; this important development gives tumors which are better characterized in terms of the biochemical defect responsible for their onset. This could become a specific target for the development of new drugs.
There is no doubt that, today, a more rational approach based on the molecular characteristics of the proteins transcribed from genes causing the stimulation or suppression of cell growth are available. Antisense oligonucleotides offer a new approach for selective interaction with specific RNA messengers, thus inhibiting the synthesis of specific proteins. Recent developments in molecular biology allow us to formulate new hypotheses about mechanisms so as to design more specific drugs for growth inhibition and chemoprevention.
Agents to counteract the toxic effects of today’s drugs are also important: antiemetics and colony stimulating factors (CSF) have reduced this toxicity and made chemotherapy better tolerated.
While all this new knowledge and developments are promising it should be remembered that drugs must reach the cancer cells, so they have to resist metabolism, inactivation and excretion in order to achieve effective concentrations. Hence we need to consider pharmacokinetics and metabolism for new classes of drugs which might be peptides and oligonucleotides. In addition it should be stressed that vascularization is relatively poor in most tumors and this may limit the availability of drugs. At the same time, however, it must be a target for drugs aiming at inhibiting angiogenesis.
Another difficulty is the heterogeneity of cancer cells which does not allow us to identify any single target. More complex strategies are required for the analysis of this heterogeneity which changes continuously in relation to cancer growth, dissemination and metastasis formation.
Finally, cancer cells become resistant to the toxic effects of chemicals. Several mechanisms of resistance such as the induction of enzymes for DNA repair, multi-drug resistance (MDR), multi-drug resistance protein (MRP) and others are known and these too offer new targets for drugs to reverse resistance.
Progression from art to science still calls for much work. As drugs become more specific we must remember the redundancy of cell functions and to find the "magic bullet", we may have to combine several drugs, in cocktails that will have to be developed specifically for each type of tumor.
______________________________________________________________________________
PRE-HUMAN STUDIES: INTERPRETATION OF THE EXPERIMENTAL DATA
J.B. SORENSEN, M.D.
Finsencenter, Copenhagen, Denmark
Introduction. The prognosis for patients with lung cancer remains grim, despite decades of clinical research, which have refined the treatment with surgery, radiotherapy and chemotherapy. Overall 5-year survival for newly diagnosed cases of non-small cell lung cancer (NSCLC) is only 10 - 15% calling for a better understanding of the biologic basis of the disease in order to utilize this to select proper treatments and to develop new treatment options (1). These humbling results have prompted studies of the biology of NSCLC in the hope that a better understanding at a molecular level will translate into improved survival and eventually to cure. The following will highlight some of the aspects of the recent advances in biology of NSCLC.
Molecular events in development. The events leading to development and propagation of the malignant phenotype are complex, multiple and interactive. These events take place sequentially and create incremental phenotypic aberrations, leading eventually to acquisition of the ability for uninpeeted proliferation, invasion of adjacent tissue and metastases (1). The current hypothesis is that at least 10 - 20 genetic mutations are required to produce a lung cancer cell from the normal one. These mutations cause activation of oncogenes (dominant cellular factors, which stimulate or predispose a cell to divide) and deletion of tumor suppressor genes (2). The genetic changes result in pathological changes which in early stages are reversible, but as progressive chromosomal changes become more complex, lung cancer develops. The development of cancer is due not only to abnormal cell proliferation with loss of growth control, but also to abnormalities in the cells intrinsic cell death programme (apoptosis).
Proto-oncogenes induce autonomous cellulare proliferation when activated to oncogenes. Activation may occur by point mutation, overexpression, or deletion of genetic material. Oncogenes evaluated for prognostic impact in NSCLC includes the ras-oncogenes, c-erb B -2 oncogene, also called HER-2 and neu-oncogene and Bcl-2 oncogene (3). These 3 oncogenes have also been evaluated in clinical trials for prognostic information in multivariate analysis, among which K-ras oncogene provide independent and significant additional information on survival in 2 out of 3 studies, Bcl-2 also carried significant information in 1 study, while c-erb B2 were without influence in one study (3).
Inactivation of genes, that normally regulates cellular growth and thereby have a restraining effect of tumor-genesis (tumor suppressor genes) can lead to uncontrolled cell proliferation. in many cases, inactivation occurs by point mutation of one allele, and, subsequently loss of an amount of the genetic material in the other allele (2). For many types of cancer, multiple mutations in both tumor suppressor genes and oncogenes are ultimately required to achieve full malignant transformation. Tumor suppressor genes evaluated in clinical trials in NSCLC include the p53-gene and retinoblastoma gene. The p53 tumor suppressor gene was of prognostic influence on survival in 5 out of 11 multivariate studies in NSCLC, while retinoblastoma gene were without influence in one study (3).
Tumour initiation. All lung cancer cells produce hormones and peptides, which can function as growth factors and generate growth loops. These include epidermal growth factor, transforming growth factor a , platelate derived growth factor, insuline-like growth factors, gastrin releasing peptide, neurotensin, and gastrin cholecystoquinin (2). Among these, epidermal growth factor have been clinically evaluated in multivariate analysis for prognostic impact in NSCLC and not found to be among the significant predictors in 2 studies (3).
New therapies. Some of the new therapies which may evolve based on our understanding of the biology of lung cancer may be the development of such modalities as prevention of cell division, immunotherapy, inhibition of angiogenesis, or gene therapy.
Prevention of cell division may be based on our increased understanding of the effect of growth factors in lung cancer. These are multiple and diverse stimulators, meaning that blocking the action of a single growth factor is unlikely to be effective. However, inhibition of intracellular signal transduction mechanisms that control multiple growth stimulating inputs offers a more realistic potential for intervention. Growth of cancer cells seems more dependent on a smaller number of intracellular enzymes (kinases) and signal transduction pathways than growth of normal cells, making cancer cells more susceptible to an inhition of a specific kinase. Recent studies have shown that activation of tyrosine kinases is an important growth signal in all histological types of lung cancer cells (4). lnhition of tyrosine kinase activity may be done by tyrphostins, some of which have high specificity against individual tyrosine kinases and show promise as anticancer drugs. Thus, in the future it may be possible to identify critical enzymes, which regulate the balance between cell proliferation and apoptosis in lung cancer cells. These tumors could then be treated with specific tyrphostin for that kinase.
Immunotherapy may take advantage of the observation that lung cancer can stimulate a humural antibody response and activate cell mediated immunity. Schuloff et al. (5) showed that the immune response could be stimulated by vaccination with tumor associated antigenes. Patients with NSCLC were immunised with autologous, cryopreserved, irradiated tumor cells mixed with BCG. The vaccinated patients had a delayed cutaneous rashion to autologous, irradiated tumor cells showing that it is possible to stimulate an immunoresponse to cancer cells. Another study by Takita et al. (6) showed that vaccination with lung cancer associated antigen and Freund's complete adjuvant in patients with stages I and II NSCLC improved 5-year survival rate from 34% in the control group to 75% for the specific active immunotherapy group.
Inhibition of angiogenesis may also lead to inhibition of tumor growth. Neovascularisation is critical for tumor growth. The cytokine interleukin 8 is secreted by lung cancer cells and induces neovascularisation, which can be blocked by an interleukin 8 monoclonal antibody (7). Another study by Varner et al. (8) used monoclonal antibodies to induce apoptosis of migrating endothelial cells. These antibodies inhibits adhesion to intracellular matrix proteins, thus preventing neovascularisation, thereby inhibiting tumor growth.
Genetherapy is the concept of correcting the gene mutations. However, it remains unclear whether correction of a single mutation in multistage carcinogenesis can significantly affect disease progression. It has been suggested that correction of a single critical genetic lesion may be sufficient to abrogate tumourgenesity in human cancer cells, particularly if the correction upregulates the activity of the apoptosis pathway (2). In a small clinical study, Roth et al. (9) used directly injection of a retroviral vector containing a wild-type p53 gene in NSCLC-patients and observed regression in 3 out of 9 patients treated. In addition, there was an increase in apoptosis in post-treatment biopsy specimens compared with pre-treatment specimens.
Conclusions. The advances in molecular biology and immunology have allowed studies of lung cancer, that have provided insights into genetic and immunologic properties of lung cancer cells. In the future, new tools for identifying biological factors, that are important in predicting the outcome or selecting treatment for NSCLC patients may be developed. Further research may identify new treatment techniques, which will increase our instrumentarium for treating this grave disease in order to ultimately improve treatment outcome.
References
1. Bishop MJ. Molecular themes in oncogenesis. Cell 1991 ;64: 235-48.
2. Sethi T. Science, medicine , and the future. Lung cancer. BMJ 1997;314: 652-55.
3. Sorensen JB, Osterlind K. Prognostic factors: From clinical parameters to new biological markers. In: Progress and Perspectives in Lung Cancer. Eds. van Houtte P, Klastersky J, Rocmans P. Springer-Verlag, Heidelberg (in press).
4. Taller A, Chilevers ER, Dransfield I, Lawson MF, Haslett C, Sethi T. Inhition of neuropeptide stimulated tyrosine phosphorylation and tyrosine kinase activity stimulates apoptosis in small cell lung cancer cells. Cancer Res I 996;56: 4255-63.
5. Schulof RS, Mai D, Nelson MA, Paxton HM, Cox JW Jr, Turner ML et al. Active specific immunotherapy with an autologus tumor cell vaccine in patients with resected non-small cell lung cancer. Mol Biother 1988;1: 30-36.
6. Takita H, Hollinshead AC, Adler RH, Bhayana J, Ramundo M, Moskowitz R et al. Adjuvant, specific, active immunotherapy for resectable squamous cell lung cancer: a 5-year survival analysis. J Surg Oncol 1991;46: 9-14.
7. Streiter RM, Polverizi PJ, Arenberg DA, Walz A, Opdenakker G, Van Damme J, et al. Role of c-x-c chemokinase as regulators of angiogenesis in lung cancer. Journal of Leukocyte Biology 1995;57: 752-62.
8. Varner JA, Cheresh DA, Integrins and cancer. Curr Opinions Cell Biol 1996;8: 724-30.
9. Roth JA, Nguyen D, Kemp BL, Carrasco CH, Ferson DZ, Hong WK, et al. Retrovirus-mediated wild-type p53 gene transfer to tumors of patients with lung cancer. Nature Medicine 1996;2:985-91.
______________________________________________________________________________
TYPE, VALUE AND LIMITATIONS OF NON-RANDOMIZED STUDIES, INCLUDING DESCRIPTIVE REPORTS AND HISTORICAL CONTROLS
C.J. WILLIAMS
Cochrane Cancer Network, Oxford, UK. E-mail:cwilliams@canet.org
Many studies in cancer care rely on a comparison of a new or experimental treatment with standard therapy. In such comparative studies the quality of the "control" is of paramount importance. This paper will review, using examples from the literature, a variety of non-randomised controls used in comparative studies. Where possible a systematic approach will be used to demonstrate problems from publication bias in obtaining controls, through the unrepresentative nature of unrandomized controls to bias in publication of the study itself. Overall, the evidence from systematic reviews of the literature show that the use of non-randomised methods in comparative trials results in a gross over-estimate of the efficacy of the "new" treatment being tested. This erroneous result is compounded and amplified by subsequent publication bias and the use of "narrative" reviews that select such studies for qualitative analysis.
Where systematic reviews of non-randomized studies have been carried out, they suggest that the non-randomized control groups fare worse than similar control groups in randomized trials, whereas the outcome in the experiment treatment arms tends to be similar, regardless of whether the trial was randomized or not.
The role of non-randomized studies in hypothesis setting and in helping in the design of randomized controlled trial will be discussed.
______________________________________________________________________________
STATISTICAL REQUIREMENTS AND INTERPRETATION OF RANDOMIZED TRIALS
P. VALAGUSSA
Istituto Nazionale Tumori, Operations Office, Milan, Italy
Introduction
A controlled clinical trial is defined as a scientific experiment that generates clinical data with the aim of evaluating one or more treatments applied to a given population of patients. The main objective of a clinical trial is to provide results which are not influenced by factors other than those under study and which are as precise and accurate as possible. For this reason, a homogeneous patient sample is usually selected, and treatment effects (the results) are evaluated on these patients. In general, "population" are large groups of people in a defined setting. A "sample" is a subset of a population and is selected from it. Clinical research is ordinarily carried out on samples. One is interested in the characteristics of the defined population, but must, for practical reasons, describe them through a sample. Whenever a clinical question is addressed, there are three possible explanations for the answer: the observation may be incorrect because of bias (selection, management and confounding biases) or chance (or random variation), or it may be correct.
For the observation to be valid, it must be neither biased nor incorrect due to chance. There are two general kinds of validity: internal validity and external validity or generalizability. Internal validity is the degree to which the results of an observation are correct for the patients being studied. External validity (generalizability) is the degree to which the results of an observation hold true in other settings. For an individual physician, it is an answer to the question: "Assuming that the results of a study are true, do they apply to my patient as well?". Generalizability expresses the validity of assuming that patients in a study are comparable to other patients.
Statistical considerations
When a clinical trial is conducted, the observed differences between treated and control patients cannot be expected to represent the true differences exactly, because of random variation in both of the groups being compared. In the usual situation, where the conclusion of a trial are expressed in dichotomous terms (the treatment is considered to be either successful or not), there are four ways in which those conclusions might relate to reality. Two of the four possibilities lead to correct conclusions: when the treatments really have no different effects. There are also two ways of concluding erroneously. In one situation, the treatment under study may be actually no better than the control, but it is concluded that the study treatment in better. Error of this kind, resulting in a false-positive conclusion, is referred to as alpha (a) or type I error. Alpha, therefore, is the error of saying that there is a difference when there is not. On the other hand, the treatment under study might be effective, but the study concluded that is not. This false-negative conclusions is referred to as beta (ß) or type II error. Beta is the error of saying that there is no difference when there is one. No difference is a simplified way of saying that the true difference is unlikely to be larger than a certain size. It is not possible to establish that there is no difference at all between the two treatments.
The probability of error due to random variation is estimated by means of inferential statistics, a quantitative science that, based on assumptions about the mathematical properties of the data, allows calculations of the probability that the results could have been occurred by chance alone. Most of the inferential statistics encountered in the current medical literature concern the likelihood of an alpha error and are expressed by the familiar P-value, which is a quantitative statement of the probability that observed differences in the particular study being considered could have happened by chance alone. It is commonly accepted that a P-value "<0.05" indicated statistical significance, without knowing whether the test has been applied correctly. The validity of each test depends on certain assumptions about the data: if the data do not satisfy these assumptions, the resulting P-value may be misleading. A discussion of how these statistical tests are derived and calculated and of the assumptions upon which they rest can be found in a number of excellent textbook of biostatistics.
The statistical power of a study (1-ß) is also determined by the nature of the data. When the outcome is expressed on a nominal scale and is described by counts or proportion of events, its statistical power depends on the rate of events; the larger this rate, the greater the statistical power for a given number of subjects at risk. If the outcome is a continuous variable (e.g., alkaline phosphates), power is affected by the degree to which patients vary among themselves: the larger the variation among patients, the lower the statistical power.
Suppose one has designed a clinical trial to compare two treatments. One is aware that random variation can be the source of whatever differences are observed. How many patients (sample size) would be necessary to make an adequate comparison of the effects of the two treatments? The answer depends on four characteristics: the difference in outcome between the two groups, alpha error, beta error and the nature of the date being studied.
Comment
Statistics is a specialized field with its own jargon (e.g., variance, regression, universe, power) that is unfamiliar to many clinicians. However, leaving aside the complexity of statistical method, inferential statistics should be regarded by the non-expert as a useful means to an end. This abstract made no attempt to deal with this subject in a rigorous, quantitative fashion. For that, the reader is referred to a number of excellent textbook of biostatistics. A clinical trial requires countless resources in terms of time (planning, conduct, analysis) and people (investigators and patients). If the study is not planned, performed and analyzed correctly, it will only create confusion and uncertainty. A medical oncologist is not required to apply statistical tests and make calculations. Nonetheless, understanding the principles of statistical inference will allow him/her to improve the quality of the design and conduct of a clinical research and to be able to make critical judgements on the results reported in the medical literature.
Suggested readings
1. Bailar JC III, Mosteller F (ed.). Medical Uses of Statistics. Waltham, Massachussetts, NEJM Books, 1986.
2. Chalmer TC. A potpourri of RCT topics. Controlled Clin Trial 1982; 3:285-298.
3. Feinsten AR. Clinical Epidemiology. The architecture of Clinical Research. Part 2: Outline of Statistical Strategies. WB Saunders, Philadelphia, PA, 1985.
4. Friedman LM. Furberg CD, DeMets D:. Fundamentals of Clinical Trials, 2 edn, John Wright PSG, Littleton, MA, 1985.
5. Peto R, Pike MC, Armitage P, et al. Design and analysis of randomized clinical trials requiring prolonged observation of each patient. Br J Cancer 1976; 34: 585-612; 1977; 35: 1-39.
6. Pocock SJ. Clinical Trials. A Practical Approach. John Wiley and Sons, New York, 1983.
______________________________________________________________________________
META-ANALYSES OF RANDOMISED CONTROLLED TRIALS
J. F. TIERNEY, L. A. STEWART
MRC Cancer Trials Office, Cambridge, UK
Meta-analyses are key elements in the objective and systematic evaluation of health care interventions. Their value is increasingly acknowledged by health care professionals, policy makers and researchers. Faced with unmanageable amounts of information, they require reliable summaries of research on which to base clinical and policy decisions.
The most reliable assessment of any medical intervention is provided by randomised controlled trials (RCTs). However, trials are rarely large enough to detect moderate but potentially worthwhile differences. In cancer research, breakthroughs are rare and we may generally anticipate that new therapies will result in only modest improvements in survival, perhaps of the order of 5 - 10%. Such benefits may be extremely important to individual patients and in common diseases, such as lung cancer, have considerable impact on public health.
However, to detect a 10% absolute benefit in survival from 35% to 45% would require that a total of 1000 patients were randomised and to detect a 5% benefit would require 3500 patients (90% power, 5% significance level). Thus in any group of trials addressing similar questions, many will be inconclusive, while by chance alone a few may demonstrate statistically significant ‘positive’ or ‘negative’ results. However, combining these results in a systematic review or meta-analysis may give sufficient statistical power to reach a clear answer. Furthermore, by reviewing and analysing all relevant randomised evidence, the results will give the most reliable estimate of effect, rather than emphasising individual trials with particularly striking results.
In any systematic review or meta-analysis, biases should be minimised and as much as possible of the randomised evidence addressed. Those meta-analyses which collect, check and analyse centrally "raw" individual patient data (IPD) perhaps make the greatest efforts to achieve this. Advantages of the approach over other forms of rmeta-analysis stem from the increased accuracy and updating of material and the flexibility and extent of analyses that can be done. For example "time to event" analysis of survival data can only be done with IPD.
It is important to identify all RCTs. ‘Positive’ trials are more likely to be published than ‘negative’ or inconclusive ones (publication bias) and particularly striking results are perhaps more likely to be published by high impact journals indexed by bibliographic databases. Furthermore, such databases do not cover all medical journals and indexing is not always accurate. Thus a simple electronic literature search is likely to result in a sample of trials biased towards the positive. For most systematic reviews the quest for trials will therefore additionally include searching appropriate journals and meeting abstracts by hand and consultation of trial registers. IPD meta-analyses always involve direct contact with trialists whose knowledge can be enlisted to help identify further trials, particularly unpublished ones.
All randomised patients should be analysed according to the therapy they were allocated, irrespective of whether or not they received it (intention to treat). Not all trials do this with patients excluded for a variety of reasons. If exclusions are related to treatment then this could bias the results of a trial. The aim of IPD meta-analyses is therefore to seek and analyse information on all randomised patients.
It is also important to include only randomised data; the problems of non-randomised studies are well known and there is evidence that trials using the least secure methods of allocation show the largest treatment effects. Trials which allocate by pseudo-random methods, such as date of birth or freely accessible lists of random numbers, could be biased. IPD meta-analyses have the advantage of being able to remove any non-randomised patients who may have erroneously been included in a trial, perhaps as part of a pilot phase testing the acceptability of a new treatment.
Owing to pressure to present results quickly, trials are often published soon after closure and few re-publish with long-term follow up. In this way trial results become fixed in time. Meta-analysis therefore provides an excellent opportunity to seek updated follow up and provide a long-term picture on the effect of intervention.
Meta-analyses which give such a picture based on updated, carefully checked IPD from all relevant RCTs provide a reliable assessment of treatment effect to use as part of the clinical decision-making process.
______________________________________________________________________________
TAKEN FOR GRANTED THE PROCESS OF EVIDENCE FORMATION, WHAT SHOULD WE COMPARE TO SAY THAT TREATMENT 'A' IS BETTER THAN TREATMENT 'B'?
M. TAMBURINI
Division of Psychological Research, Istituto Nazionale Tumori, Milan, Italy
A document on Outcomes of Cancer Treatment for Technology Assessment and Cancer Treatment Guidelines (adopted on July 24, 1995, by the American Society of Clinical Oncology, and published in: J Clin Oncol 1996, 14:671-9) affirms: "Survival, whether measured overall, disease-free, progression-free, or event-free, is the most important outcome in cancer treatment. Nevertheless, survival alone is not sufficient; the quality of survival and cost of maintaining or improving it must also be assessed".... "The choice between alternative treatment approaches often involves a trade-off between length and quality of life; survival alone may not answer the question of whether gains in survival justify the toxicity".
Whereas for a definitive cure of a disease it is acceptable to pay the price of undesired side effects of the treatment, especially when it is of a transitory nature, this may be not justified when the period of life induced by the therapy is short and of poor quality.
Definition of quality of life is still particularly difficult. In its widest concept it is identifiable with that of happiness. The entire personality, taken as the result of experience and genetic patrimony of the individual, concurs in the subjective perception of a good or poor quality of life. The self-security, emotional conflicts, personal ideals and aspirations, and the degree of tolerance to frustration all influence in a determinant manner the perception of well-being or malaise. It is well known that two individuals confronted by the same situation can react in a completely different way. The birth of a child generally gives rise to intense moments of joy, but the opposite can happen for one who does not desire such an event. Death of a family member usually causes great sorrow, but it can favor a contrasting emotion for one not effectively linked to the deceased and for some time waiting for the inheritance. It is often not the event itself as much as the significance of the event that can be expressed by positive or negative emotional valor.
While waiting for a stricter definition of quality of life, for about 10 years researchers have been oriented towards a working definition, identifying four fundamental dimensions: functional status, physical symptoms, emotional and cognitive conditions, and social relations.
Detection of the quality of life is a complex operation. Physicians and nurses tend to underestimate the functional status and some physical symptoms (such as pain and dyspnea) and to overestimate the degree of psychological discomfort (i.e., anxiety, depression and distress) (Sprangers MA, Aaronson NK. The role of health care providers and significant others in evaluating the quality of life of patients with chronic disease: a review. J Clin Epidemiol 1992; 45: 743-60). The logical consequence of an underevaluation is an undertreatment.
In the last twenty years there have been numerous changes in quality of life assessment: the first and perhaps most important was the passage of an evaluation based on the impression of the physician to one in which the judgment is given by the patient. This change was followed by greater attention being given to multidimensionality, the modality of a structured assessment, the identification of the properties of measurement instruments, and the predisposition of specific modules for specific diseases or treatments, to then reach multilingual adaptation of the evaluation instruments.
Table 1 reports a list of the most multidimensional quality of life assessment instruments used in cancer.
Instruments using specific modules or items for lung cancer assessment are: the EORTC Quality of Life Questionnaire Core 30 Items (QLQ-C30) with lung cancer module QLQ-LC13; the Functional
Assessment of Cancer Therapy - Lung (FACT-L); and the Lung Cancer Symptom Scale (LCSS).
|
