Introduction

The ethical principle underpinning medicine is ‘primum non nocere’—first, do no harm. As research leads to an improved understanding of disease processes however, it can be difficult for medical practice to keep up with changing knowledge and adhere to that concept. A recent publication from Karagiannis and colleagues has attracted much interest, especially among the patient community, due to its finding that neo-adjuvant chemotherapy appears to cause micro-environmental changes linked to metastatic breast cancer progression [1]. The specific focus of this paper is on a set of tumor cell and stromal alterations that promote the process of breast cancer cell intravasation. It highlighted how the invasion of cancer cells into the lymphatic system or blood stream is altered by chemotherapy. The general concept of therapy-associated metastasis is not new and has been reported previously for taxane therapies [2]. Indeed all standard therapeutic modalities—surgery, radiation, chemotherapy and molecularly targeted agents - have been linked in the research literature to increased markers of metastatic progression in different cancer types (reviewed in [3]) and new studies regularly emerge [4, 5]. Nevertheless, these same therapies are widely used clinically to treat patients at risk of metastatic disease, leading to questions of whether harm is being done. This dichotomy has rightly resulted in some recent confusion, particularly amongst the metastatic patient community. Below, we discuss the patient perspective then offer reasons why research findings, whether positive or negative, should be interpreted cautiously and cannot typically be immediately translated to changes in clinical care. Nevertheless, the purpose of researching how therapies may influence metastasis is to ultimately improve standard of care and benefit to patients. It is thus important for clinicians and basic researchers to work together to design appropriate clinical trials that will provide answers for patients looking for therapies that are safe and effective both short- and long-term. The paper from Karagiannis et al., in fact aims at providing a future solution to this problem.

The patient perspective

When a new scientific article comes out, most patients rely on mainstream journalism (TV news, newspapers, magazines, etc.) to summarize the science and explain the main findings. The majority of scientific articles are not accessible to the patient community directly because they are not published as ‘Open Access,’ which can cost approximately $3000–$3500 per article. Unfortunately, journalists covering scientific articles generally don’t explain the nuances of the science involved. Instead, they write a flashy headline designed to get ratings and page views, a headline that too often is an inaccurate or inadequate description and interpretation of what the data in the article is actually showing.

It is vitally important for patients to discuss mainstream cancer headlines they read with their oncologist, who can accurately explain to them what the article shows, offer an appropriate interpretation, and explain what the findings could mean for an individual patient’s treatment. Each patient’s cancer is different, and each patient’s treatment is individualized for their particular cancer. A patient’s oncologist is hands down the most knowledgeable person regarding that patient’s cancer, and what the risks and benefits of individual treatments are for that patient. It is risky for a patient to reject their oncologist’s treatment suggestions based on what they read in the mainstream news. Patients and their oncologists need to work together to talk about the latest science and the patient’s individual needs, with the goal of keeping that patient alive and feeling as well as possible.

Putting research findings into context

Oncologists should be the leading resource for patients trying to interpret translational science results and apply them to their own care. There are also some important limitations of several common basic research and clinical trial attributes and techniques that all can keep in mind to more accurately interpret the potential clinical impact of research findings or trial outcomes in general. There are several differences between bench research, clinical trials and ultimately ‘real-world’ outcomes. It is incumbent upon researchers, whether basic, translational, epidemiological or clinical, as well as clinical care providers including those disseminating information directly to patients to acknowledge these differences and understand the limitations they impose.

  • Animal models It is recognized that all pre-clinical models have limitations that can contribute to the failure of novel therapeutic agents in clinical testing [6,7,8,9]. These limitations also apply to predicting therapy complications for many of the same reasons. With animal models, researchers generally use a homogenous subject group where the mice are of the same age, genetic background and have been exposed to the same environmental factors. In contrast, human patients are extremely diverse when considering these parameters. Additionally, cancer models in animals are often generated by either injecting a large number of already mutated cancer cells all at once, or by expression of a strong oncogenic gene, which in most cases allows rapid cancer development in a short period of time (weeks to months). These events happen in the setting of an otherwise healthy animal (except perhaps for being immune-compromised in tumor transplantation studies). In people however, cancer can develop from rare mutated cells over many years. Progression is influenced by underlying complex genetic susceptibility traits and can be exacerbated by chronic diseases such as inflammatory conditions or other stresses on general health such as obesity, alcohol-intake or smoking. Thus, findings that are highly significant and consistent in model animals, become much less obvious and more difficult to interpret when there are multiple other variables in human patients. Improvements in tumor modeling in animals can alleviate some of the limitations and ultimately should increase the predictive value of preclinical evaluation. Such improvements include for example using multiple patient-derived xenografts to add heterogeneity, using animals with already established metastatic disease prior to the initiation of therapy and/or using several different types of well-established models—e.g. tumor implant, carcinogen-induced or genetically-engineered models.

  • Use of clinical specimens A strong component of the study from Karagiannis et al. was their analysis of clinical specimens to show similar changes as observed in the animal studies [1]. Of course, there are limitations regarding the type of questions that can be asked using clinical specimens. Specifically, in such samples it is very difficult to determine cause versus effect directly; instead investigators must look for associations. The strongest data generally come from longitudinal studies where repeated sampling of the same individual (e.g. from multiple blood draws, biopsies and/or surgical resections) allows determination of when particular changes became apparent. These kinds of studies are challenging to complete partly because repeat biopsies cannot always be justified and partly because patients may drop out of a study before its completion for a variety of reasons. The result is a negative impact on sample size and subsequently the power of the study to draw robust conclusions. Even with longitudinal studies, the endpoint is often relatively short-term constituting measurements that can be made with the samples collected. They often do not extend past the 5 years or longer that may be necessary to see whether there is a long-term impact such as development of overt metastases or even survival.

  • Clinical trial endpoints The use of chemotherapy as a neoadjuvant therapy (i.e. treatment given prior to the main therapy, usually surgery) in breast cancer patients at high risk for progression was initially tested and showed to be beneficial in relatively small trials [10, 11]. More recently, larger clinical trials have continued to support these findings [12, 13]. The trials showed that in particular populations of high-risk breast cancer patients, treatment with drugs such as paclitaxel prior to surgery was associated with an improvement in surrogate endpoints, for instance ‘progression-free survival’ or ‘pathological complete response’. Other studies in breast cancer have shown improvements in disease-free survival for neoadjuvant use of radiotherapy [14] or hormone therapy [15]. Often, results from these trials report patient outcome at 5–8 year time points by which time disease recurrence is assumed to occur. Yet, when more data over a longer term is available, it seems that there is not always a corresponding increase in overall survival, although this may be dependent on cancer type and aggressiveness [16,17,18,19,20]. A separate study compared ‘real-world’ outcomes derived from Surveillance and Epidemiology End Results-Medicare data with either overall survival or surrogate survival endpoints from 21 Phase III trials across a range of cancers. This study found agreement when overall survival was the trial endpoint. However, efficacy was reduced when surrogate endpoints were used [21]. Paradoxical findings of improved disease-free survival without affecting overall survival provide rationale for examining untoward effects of neoadjuvant therapy. Nevertheless, it is very important to note that therapy does not appear to lead to a decrease in overall survival either [12, 14]. It is also important to realize that using only the metric of increased overall survival for multiple cancer types could impede rapid availability of new, efficacious drugs that may bring significant benefits to patients, such as improved quality of life. Thus, the balance between using an endpoint that suggests some shorter-term benefit yet no improvement to long-term outcomes, versus only approving agents that increase overall survival significantly is a still-evolving issue [22, 23].

  • Patient stratification—the use of molecular markers In breast cancer, ‘precision oncology’ is not an especially new concept given that treatment decisions are already made based on the expression (or lack thereof) of hormone receptors (estrogen and progesterone receptors, ER and PR) and HER2. There is some evidence that breast cancer subtype, based on expression of these receptors, is an important factor in the initial response to neoadjuvant chemotherapy [24, 25]. However, for the longer-term outcome of distant recurrence, tumor size rather than subtype appeared to be a more powerful predictor [26]. More sophisticated stratification of patients may be necessary to determine who is likely to benefit from therapy in both the short and long term [27]. For example, a study examining the molecular profiles of tumor specimens from 861 breast cancer patients on a clinical trial concluded that use of PAM50 and triple-negative genetic subtyping could identify distinct prognostic subsets that had different outcomes compared to overall population [28]. The clinical trial from which these specimens were analyzed (USO 01062) was one testing the efficacy of adding capecitabine to anthracyline/taxane adjuvant chemotherapy in high-risk early stage patients. The conclusion was that in the large, heterogeneous study population, there was no demonstrable benefit of the combined treatment. However, if this trial had instead only included basal-type triple negative cancers there may have been a beneficial outcome. Thus, in trials where stratification is not considered, therapeutic approaches that work well for a small subgroup of patients may not be picked out against the larger pool of patients showing no benefits. Where bench studies have identified factors that may predispose certain subtypes of cancer to a better or worse response to any therapy, it would be valuable to examine which of these factors could be included in subsequent clinical trials of that therapy.

A myriad of other factors can impact reported trial outcomes such as the initial disease burden in patients enrolled in a trial, dose and timing of the administered therapy [29] and even (for progression-free endpoints) bias based on timing of evaluation [30]. Given the complexity involved in determining what therapies should be used when, and with what patients, it is remarkable that the system works as well as it does. However, the continued evaluation by basic and clinical researchers of how drugs behave in more sophisticated animal/cell models as well as in real-world clinical settings provides much-needed feedback to improve the process. Thus, findings such as those reported by Karagiannis and others offer insights into potential effects that should now be assessed using clinical outcome data to ensure that patients are offered treatments that provide benefit but that also do no harm.

Finally, patients recognize the need to make treatment decisions based on scientific evidence, but it can be frustrating and frightening to realize there is so much about metastatic cancer treatment that is still unknown. While patient choice is an absolute right, treatment decisions should be made with a good understanding of risk and benefit. Patients should continue to rely on their oncologists to help them interpret the relevance of an individual study in the overall context of current evidence. Unfortunately, patients too often hear “we just don’t know the answer to that.” The Karagiannis study underscores what patients already know—that more research is desperately needed.