The growing intersection of pharma and technology
Joe Lukasik is the director of Technical Services for Linguistic Validation at RWS Life Sciences. He has worked in the localization industry since 2000 in a range of areas including electronic clinical outcomes assessment (eCOA) production in partnership with the world’s largest pharmaceutical and eCOA providers. He holds degrees from the University of Rochester and the University of Michigan, and lives near Boulder, Colorado.
he last few years have seen considerable growth in pharmaceutical clinical trials around the world. Worth an estimated $44.2 billion in 2018, according to Grand View Research, the global clinical trials space is expected to grow at a compound annual growth rate of 5.7% and reach a market size worth $68.9 billion by 2026. Among the most compelling factors behind this rapid growth — and driving the potential for pharma — are globalization and the evolution of technology.
It used to be that drug companies would release new products in countries that offer the best return on investment: the US, which takes the biggest share (currently 40%) of the pharmaceutical market, followed by Japan and, at a close third, China. Globalization expands the possibilities, however, as drug companies have recognized new opportunities to get regulatory approval for sales in wider geographic regions — most recently including Eastern Europe, India, Southeast Asia and Africa.
New technologies have opened these markets by solving the challenges that plagued them, for both pharma and its global partners. But those challenges were not insignificant.
When it comes to localization, obstacles arise when dealing with countries we haven’t dealt with before, within narrower time frames as drug companies race to get their therapies out sooner. The geographic footprint of clinical trials is also far outpacing the resources available to do translations, reviews and other language-related tasks, as well as linguists’ access to modern technologies.
Indeed, until recently, pharma has lagged behind on employing localization and content reuse/management technology as a means of accelerating growth. A life sciences localization industry inside joke used to go something like this: “the customer’s idea of version control was to create a PDF and throw the source file away!” Though this is a bit extreme, there is a bit of truth here: concerns about version control, specifically in maintaining exact versions of documents that were already filed with overseas regulatory bodies, tempered the willingness of many companies to adopt technologies that allow a more agile flow and reuse of content. Pharma companies preferred to retranslate from known source versions using expert human translation. The additional expense was worth mitigating perceived risks.
What’s changing, and rapidly, is the attitude toward these legacy manual processes. As pharma companies aim to secure drug approvals as quickly and safely as processes allow, they face pressure to find ways to speed up processes to get new therapies in the hands of patients faster and see more rapid worldwide market results. To do this, technology has become essential.
The rise of technologies
About a century ago, Henry Ford decided he wanted to make more cars faster. His solution was to invent the assembly line. In 1913, the first moving assembly line was installed for the mass-production of an entire automobile, reducing the time it took from more than 12 hours to two and a half. But to support the process, he needed to invent new technology as well.
Think about this from the context of releasing a brand-new drug. Markets are closed to new drugs until they are proven safe through clinical trials and certified by the local regulatory body governing that market. So, to benefit more patients sooner, pharma companies have realized that the process of drug regulatory approval should happen in parallel across more countries.
A clinical trial in a single country generates an impressive volume of data in the form of case report forms, lab reports and various other data collection methods. Now imagine the total volume of incoming data as pharmaceutical companies run trials in as many as 40 to 80 different countries. Despite the challenges of dealing with the growing volume of data, this presents a unique opportunity for pharmas to view trends across a much wider and diverse population than ever before. This in turn is an opportunity for language service providers to develop service offerings that include translation to a common language and data normalization that would allow efficient analysis of large corpora of data.
It is also logical that expanded clinical trials across many geographic areas require translation from a single language into many, since local regulations require test subjects to receive information in their own dialects that is clearly understandable according to their education level and cultural norms. This is difficult enough when translations are created to educate or inform test subjects about the trial in which they are participating; it elevates to another level of complexity when the translations used for data collection can impact new drug evaluation and approval. For the latter, a process called linguistic validation is used in which translations progress through a special process to ensure complete and consistent comprehension of the questions, regardless of the language. Any divergence of data due to a poor translation can delay the evaluation and approval of a new drug release.
It is within this expanding ecosystem that technology comes into play. We can boil its role down to four key factors:
Technology accelerates the ability to inform patients quickly.
Technology provides the fastest method for collecting information from patients.
Trials, supercharged by data collection devices like phones and tablets (and soon to come, smartwatches), are generating more data than ever before.
Technology is the only way to process and analyze this staggering volume of data.
It’s time to take a closer look at trends in the pharmaceutical industry and related opportunities for technological innovation.
Greater volumes of content to inform patients
The rise of clinical trials generates unprecedented amounts of supporting content. Global clinical trials require multilingual content for marketing, training, pharmacovigilance, site- or patient-facing materials and more, some of which need not only translation, but also certified processes or scientific validation as mentioned above. Tightening compliance standards and regulations, ranging from global guidelines to country-specific laws, have also generated more content.
Yet as the rate of production rises, so does the bar for translation quality. Accurate translations are crucial for maintaining trust in doctor-patient relationships, for patient safety and education and for appeasing drug regulators. The regulatory fallout from the smallest error is immense: if halted by the FDA, clinical trials sink billions of dollars and damage the delicate trust among regulators, clinics and patients.
So, as companies plan to release new drugs and products in more countries at once, there is often a need to produce translations at volume, in many languages and at high speed. In the medical device industry, for example, we would need to translate an entire catalog of components, from filters to chords, each with its own description. This is a perfect application for machine translation (MT). Pharma, too, is looking into MT as a means of back-translating data generated in many countries to one language, which can be more easily analyzed for wider trends during clinical trials, or after in the form of doctor or patient/consumer feedback as part of pharmacovigilance or customer service chat exchanges.
But what about quality? In an environment as fraught with risks as pharma, it’s important to apply disciplined review strategies to output from automated tools like MT, rather than rely on the raw output without human input. MT is generally useful only for less mission-critical applications, such as back-translation in pharmacovigilance, the translation of informational content or perhaps customer service chats. For translation of data collected in clinical trials — such as patient reported outcomes (PRO), which set critical expectations for patient comprehension and integrity of data collected — MT remains high-risk. After all, pharma is an industry with a high cost of failure and where patient safety is at stake.
The increasing volume and complexity of patient data collected
As mentioned earlier, drug companies have made strides to maximize the initial release of their products. One way has been to rethink patient data collection.
What used to be a standardized test-style form on a clipboard has now migrated to electronic clinical outcomes assessment (eCOA) software that runs on a mobile device or tablet, revolutionizing our approach to clinical research. On paper, COA questions were extremely focused and limited: a heart disease trial might ask questions around relevant factors like breathlessness or fatigue, but would miss opportunities to understand patients’ overall well-being like depression or quality of sleep. While the key questions remain the same as part of scientifically validated data collection instruments or questionnaires, by using an electronic format, clinical investigators have additional ways to collect more data, such as through patient diaries and other patient-generated information. eCOAs also give patients a chance to submit other types of free-form information that might uncover side effects unanticipated by the clinical investigators.
In addition to the data coming in through outcomes assessments, pharmas are suddenly getting much more information back from clinical trials as they run more on a wider scale. And in the future, as we’ll discuss shortly, we’re likely to see yet even more data come in from a wider array of data collection devices.
In the meantime, information is reported in three main ways: by the patient through PROs and patient diaries, by the trial test sites through case report forms (CRFs) and through blood tests, scans and questionnaires completed by the clinician. Streaming in from multiple sources, the data must be normalized in a way that can be easily and consistently analyzed across countries. First, if collected in several languages, the data might need to be translated into one language. Then it’s the language service provider’s (LSP’s) job to ensure consistency of terms, convert number formats into the same units and perform any other checks required to prepare the data for apples-to-apples comparison and trend identification.
Can this normalization (through translation) be done using MT? In some cases, yes. Another challenge, however, is that some information coming in from doctors or patients is handwritten. Physicians’ scrawl remains the gold standard for illegibility until we see more improvements in handwriting recognition technology. The use of MT also depends on the context and format. A handwritten serious adverse event must be submitted for immediate translation by a human, while electronic CRFs might be machine translated first and then reviewed or fully back translated by an expert clinical linguist.
Increasing the complexity of how we deal with data is the entry of companies like Apple, which plans to add another data stream to the mix: continuous patient telemetry, as we’ll discuss next.
New patient-facing technologies
There was once a firm division between the medical device and pharma worlds — between the drugs we take and the devices we use. Now, the latest buzz is around tech giants investing in supporting global clinical trials by finding ways to combine medical devices with consumer products such as wearables. For example, an insulin pump might one day pair with an Apple Watch to remind patients to replace their vial of medication. At a minimum, smartwatches will be able to send telemetry (heart, sleep and activity data) directly to the clinical investigator.
What we’re finding, then, is that patient telemetry can constantly flow back to the investigator. Before, interaction with a patient during a clinical trial was more point-to-point: the patient takes the drug, fills out the diary, fills out a PRO after coming in for tests and repeats this process until the trial is over. Information flowed two ways, but certainly less granularly. Now, there is no need to wait for transactional exchanges for certain types of information.
For patients, the benefits of this are enormous. Drugs are intended to improve patient outcomes in the obvious sense, of course. But the more granular the data, the clearer the picture of outcomes both positive and negative — that is, the extent of a drug’s safety and efficacy during testing.
There is also interest in creating patient-facing technologies like apps that have a dual purpose: convenience for patients and insight for pharmas. Apps that allow patients to enter information voluntarily could collect critical data about patient experiences with an approved drug, their progress with a life-saving therapy or over-the-counter drug use. Websites could collect more data from patients after a trial or release of a drug as part of end-to-end patient wellness initiatives.
This is all to add more data (and excluding patient telemetry, more content to translate) to an already swollen data pool. If we’re going to work faster and smarter to keep up with the growth of clinical trials and the speed at which drugs go to market, there must be a way to parse and analyze the data. Artificial intelligence is one solution that benefits the pharma and localization industries alike.
The use of AI to seek patterns in data
AI is changing health care in many amazing ways. For example, by looking at dozens of x-rays, AI can quickly draw patterns and diagnose problems that are invisible to the naked eye. A study published this January by the University of California San Francisco combined neuroimaging with machine learning to detect Alzheimer’s years before diagnosis. And in a February press release, Google announced a patent for its new deep learning algorithm, which might be used to predict serious adverse events before they occur in a clinical trial.
The biggest opportunity here for pharma in particular is to use AI to analyze the ever-growing corpus of data generated during clinical trials (and after, when comparing data sets between clinical trials or even trials from different drugs) and to create what you might call a 360-degree view of drug efficacy: what the drug is designed to do, plus any unanticipated side effects. For example, pharma only recently found connections between dementia and Benadryl, a drug that’s been used for decades. This insight is good news, especially for patients who use drugs after approval under less supervision than they’d receive during trials. With the patient wellness movement, sprung from the realization that preventative care and a healthy lifestyle cut medical bills in the long run, pharma’s recognition of responsibility for long-term users of their medications and continued data collection by physicians during routine physical exams, pharma can now continue to collect data for extended periods after clinical trials and approval. AI will be able to help spot trends too obscure for most humans to notice. Imagine its potential on a global scale!
In the past, regardless of location, the volume of data collected in a clinical trial was limited to the volume that could be viably processed and analyzed by human analysts. As mentioned, patient questionnaires were generally closed and curated experiences with only multiple-choice options, which was the only way human analysts had a chance.
But with AI, we could analyze free-form responses. ePRO designers can now create text fields that allow patients to describe their conditions in their own terms and report on side effects or experiences that were excluded from the more focused COA questionnaires. This won’t be a validated methodology of data collection, but it can at least raise flags about unanticipated side effects and trigger further investigation through forms like PROs. AI thereby allows for more context as a triangulation between telemetry and free-form data.
Added to this, AI can help speed up analysis of indicators of a patient’s overall health and quality of life, well beyond PROs and visits to the trial site clinic. Efficient analysis, too, means that trials can be conducted in even more countries and regions. With AI, analysis and insights from huge data sets will quickly become the norm. Soon, no data will be left behind.
Even more interesting is that the AI-powered, mixed-device world brings more business opportunities for localization suppliers. First is the sheer volume of content to translate just to support device use, not to mention data from other devices, the additional free-form content and, in some cases, the data collected after clinical trials.
AI can also benefit the translation process itself — with better MT, obviously, but also the ability to move further away from manual production (using metadata to automate which models to apply to which translation problems, for example). Localization providers might also help by offering services to normalize multi language free-form data into one language to facilitate AI analysis and help pharma develop a more agile pharmacovigilance model.
What does the future hold?
There’s no blueprint yet for how to leverage the full potential of technology in the life sciences industry. It’s particularly challenging to do since new technologies are popping up all the time.
For example, patient-facing technologies must rely on the patient for data accuracy. In the future, we will turn to nanotechnology, which would report telemetry more reliably from inside the body, a development that promises to be huge. I can even imagine a Star Trek-like future in which we use hand-held devices to reveal all the mysteries of the body, or syringes to synthesize any cure on the spot, or seamless teleportation of whole new organs into bodies.
How we apply these potential technologies could take unexpected directions, but given the ever-accelerating progress in other areas — robotics, machine-neural interfaces, genetic engineering and more — we can expect to see developments that were unimaginable only a few years ago
No matter what happens, one thing will remain constant: the need to prove that treatments are safe and effective. That takes data. And on top of using AI to aid in analysis, we’ll need new ways to organize and access the data as it continues to grow.
When humans can’t possibly handle all the work, how can we use AI in localization? For example, what if we could apply a preventative rather than corrective approach to dealing with translation errors? The focus in our maturing industry will be to continue monitoring, automating and optimizing process design — including that of translation — as volumes and content types explode, paving the way for a new generation of AI capabilities. Indeed, leading college programs such as the MA in translation and localization management at MIIS, which was recently reclassified as a STEM degree, are now creating curriculum in categories such as automation, data science and artificial intelligence.
There will always be more demands on localization to get products to more markets quicker, many of which are unique to the life sciences industry. On the other hand, smarter processes will enhance the efficiency of LSP services to the life sciences industry. We’re looking at a future in which localization doesn’t just help sell or secure approval for drugs — it also aids in collecting quality data that accelerates the improvement of patient outcomes. And this, ultimately, is the best outcome we could hope for.