In recent years, dozens of large corporations and tech start-ups announced that they would deliver self-driving cars by 2020, thanks to the power of advanced analytics. It seemed like a daunting but plausible goal, as newer models already included semiautonomous features such as adaptive cruise control and driver-assist parking. But over time, we discovered, based on the massive amounts of data, machine-learning models, and expert engineers required, just how complex it is to build a fully autonomous vehicle. Nevertheless, while no company is producing self-driving cars on their assembly lines yet (due to the continual pushback of estimated delivery dates), it is only a matter of time.
In oncology, the story is much the same. We have read the exciting headlines predicting that big data and advanced analytics will transform cancer care and research and that all patients, thanks to machine learning, will soon receive personalized treatment plans. But progress has been arguably slower than initially envisioned.
However, without a doubt, important advances are being made. Researchers have built a model that can accurately predict molecular subtypes in head and neck cancers using radiomics, a machine-learning technique that can help identify, classify, and monitor solid tumors from CT, MRI, or PET scans, rather than relying entirely on radiographers and often-painful biopsies. Elsewhere, machine-learning models have outperformed board-certified dermatologists in identifying melanoma through image recognition.1
Yet most pharmaceutical companies still do not systematically apply advanced analytics to their work in oncology, even though it can deliver value at every stage of the product life cycle, from research and early development to market access and commercialization (see sidebar, “The value of advanced analytics in oncology.”) They do not, for instance, routinely analyze real-world evidence to identify additional indications for existing therapies—analysis that could prove particularly helpful in treating rare cancers where it is difficult and time intensive to find patients for a trial. And they are unable to use data to understand the full network of a patient’s care team—specialists, primary-care providers, and sites of care. At a time when the COVID-19 pandemic has so hindered cancer care, think of the advances that might be made if companies could use such data to help oncologists keep abreast of the often-bewildering array of new combination therapies and targeted treatments, ensuring that patients receive the best standard of care based on their history, genetics, and biomarkers.
Lack of talent is one reason why pharmaceutical companies are not making more progress applying advanced analytics to oncology—data scientists with expertise in oncology are few and far between. Lack of data is also an issue. But perhaps the biggest impediment is a culture that is suspicious of the kind of experimentation inherent in developing advanced-analytics capabilities. As a result, many companies are simply postponing a steep learning curve, as advanced analytics will, without doubt, come to play a huge role in cancer research and care, as well as pharmaceutical companies’ success in oncology.
Paving the way for advanced analytics in oncology
A management-aligned technology road map that makes clear the expected impact, timeline, and investment required to deliver on priority initiatives is the starting point for any successful advanced-analytics strategy, regardless of sector. Advanced analytics can cover a range of approaches, from statistical models to artificial intelligence (AI) and the deployment of machine-learning tools such as the neural networks used for image recognition. Analytics models can be built without a clear, strategic road map, but they won’t necessarily give the business a leading edge or, of critical importance in the case of pharma companies, help patients. A strategic technology road map is the first of our six-component framework for successfully integrating digital and analytics technologies to create value, the others being talent management and planning, an agile delivery model, data-strategy and data-management capabilities, technology capabilities, and the implementation of measures to drive adoption of a new operating model (Exhibit 1).
While all are key, we focus here on the elements that can prove particularly important in overcoming challenges in oncology, namely, building the right talent within the right organizational structure, kickstarting the program by leveraging existing data assets first, and embedding a culture that will speed adoption.
Build the right talent in the right organizational setup
Large, established pharmaceutical companies are not short of data scientists, but most of them work on commercial analytics, executing repetitious analyses such as performance reporting. If companies are to capture the full power of analytics, they will need to consider devoting specialist resources to different therapeutic areas (TAs). This is particularly important in oncology, which poses unique analytical challenges, such as integrating pharmaceutical and medical data, translating clinical markers, and building complex lines of therapy.
Finding such expertise is a tall order. It is rare to find talent with a strong technical background as well as a medical understanding of a disease area, the ability to link clinical guidelines to all manner of data in the patient journey, and the business sense to recognize and prioritize the most valuable insights.
Some companies might therefore turn to data-science companies and contractors to resource analytics projects and deliver insights fast. But if the aim is to establish a competitive advantage through advanced analytics in oncology, companies will probably need to build analytics excellence in-house. This, in turn, will require a strong talent-recruiting strategy that differentiates the cutting-edge role of analytics in the pharmaceutical industry—particularly oncology—and ensures well-defined career paths and growth opportunities.
Another question to settle is whether these new data scientists should sit in a centralized analytics center of excellence, which provides platform services across the organization, or within the oncology division itself. Centralization versus decentralization is a much-debated design choice in all sorts of companies. Some opt for a centralized model to help ensure resources are used efficiently and data scientists learn from one another. Some prefer decentralization to keep data scientists more attuned to the needs of the business. And some strike a balance with a hybrid model. The size of the oncology portfolio could guide the choice. With a single oncology drug, the best option might be for analysts to sit within the business unit. With a larger portfolio, there could be benefits to them sitting within an oncology cell in a centralized analytics function to share and scale their knowledge.
Kick-start the program with existing data assets
Data needs for oncology use cases are more specialized than those for other TAs. For example, mapping a patient journey containing combination chemotherapies requires integrating medical, prescription, and laboratory-results data sets, while identifying later lines of therapy may require three or more years of patient-history data.
The quality, availability, and cost of oncology-specific data sets have undoubtedly improved over the past five years. Yet, there is still a long way to go. Data remain relatively sparse, and sample sizes tend to be low, particularly for biomarker data due to the collection cost, potential errors in measurements, and the risks to patients when collecting specimens such as neural tissue.2 Additionally, with the exception of electronic-medical-records (EMR) data, clinical data seldom include the outcomes of tests or procedures. Claims data, for example, exclude laboratory values and genomic test results.
Notwithstanding the pressing need, time should not be wasted waiting for perfect data. To kick-start the analytics program, pharmaceutical companies should identify and put to quick use as many existing data assets as possible—including those typically used for operations and standard reporting, which often get overlooked when it comes to supporting advanced analytics. Data on rebate cards, oncology practices and cohorts, and oncologists who have participated in clinical trials could all prove valuable.
At the same time, companies should look to acquire novel external data sets, including biomarker data and even data from digital-health monitors. Ultimately, they can combine existing and new data sets to reveal even more powerful insights, notwithstanding that new data assets can take time and most companies already have data that they consider “good enough.”
As more data is incorporated and use cases proliferate, data governance becomes critical if the data are to be usable, accessible, and secure. This should include governance of top-level issues, such as who owns the overall data strategy, as well as of tactical issues, such as how to define data fields. Take the data field for physician affiliation, for example. Oncologists are often linked to several hospitals, clinics, and offices, which means the best definition of “affiliation” would depend on which question you are trying to answer: Where are oncologists seeing patients, where do sales reps interact with them, or through which entity do they order or bill for medications?
Change the culture
Even companies that build strong data-science teams and strong data assets can struggle to get new tools adopted. The problem is pervasive across sectors, but it can be more prevalent in life sciences, including pharmaceuticals, where the prevailing orthodoxy is that data and models must be perfect before being harnessed for decision making. That orthodoxy is correct when applying analytics to some areas within R&D, but there are plenty of other areas where a less-than-perfect analytical tool can be beneficial. Moreover, given how quickly standards of care can change and the complexity of different lines of therapy, companies will struggle to get off the starting line if they wait for perfection.
Lack of transparency and training can also hamper adoption. Field teams handed a tool backed by complex models are often skeptical that it can outperform their years of institutional knowledge, particularly if the drivers of a recommendation are unclear, which is often the case with AI models. The skepticism is compounded if the tool also fails to make clear what action to take as a result of the analytics—a situation that is not uncommon.
Overcoming such issues requires a cultural change that embraces advanced analytics. Several actions can help drive one:
- Change the way teams work. Tools will only be adopted if the teams that build them are integrated, putting to use the expertise of data scientists, analytics translators, and business leaders. Translators are the bridge between the technical knowledge of the data scientists and the operating expertise of business leadership. Their role is to ensure data scientists are going after the highest priority business problem and that the business will adopt new analytical tools by explaining their output and value. Teams also need to adopt agile ways of working, rapidly building solutions and testing and learning as they go, rather than striving for the perfect answer at the outset (Exhibit 2).
- Identify quick wins. To build enthusiasm, start with high-priority pilots that are not overly complex to conduct but that might deliver important insights. In development, that could be identifying patients eligible for a given protocol in each hospital in a region or country. For medical affairs, it might be understanding oncologists’ treatment regimens in terms of dosage and duration with a view to improving them where appropriate.
- Standardize. Standardizing the way new projects are launched and their progress measured accelerates adoption across the organization, positioning them as continuous, ongoing efforts to build competitive advantage as opposed to one-off pilots.
Different companies will find themselves at different starting points, facing different challenges when considering an advanced-analytics strategy for oncology. Larger, established pharmaceutical companies will likely benefit from a wealth of internal data sets and more data scientists. On the other hand, they might struggle to shift a more deeply engrained, cautious culture. Smaller biotech companies are likely to be more agile and more attractive to data scientists and other technical experts. However, they will have fewer existing data assets; a smaller portfolio of products, each perhaps critical to the company’s success, could also make them more averse to a new, advanced-analytics-led approach because of the perceived risk.
Scarcity of data is also a challenge for companies researching and developing treatments for rare cancers or therapies with small eligible populations. Particularly sophisticated statistical techniques will be required to derive valid insights from the data.
Whatever their circumstances, however, all pharmaceutical companies will need a strategy for applying advanced analytics to their work in oncology. Failure to establish one amounts to a decision not to participate in a technical development that is reshaping cancer care. Companies have a choice of whether to keep pace with the development or lead it, but falling behind is not an option.