The Dry Eye

Holmes and Watson: Dry eye, big data and AI

A study shows the real-world use of Restasis and Xiidra 12 months after being prescribed.

Did you know that I am an author? Yup. Really and truly. Well, a couple of other folks helped out. Dr. Paul Karpecki and I were invited by the scientists at Sun Pharmaceutical and the IBM Watson team to help design and write a paper, “Real-world treatment patterns of cyclosporine ophthalmic emulsion and lifitegrast ophthalmic solution among patients with dry eye,” looking at dry eye disease treatment patterns. Everything about this project was super cool, especially for someone who lives in the post-drug development treatment space like me.

Darrell E. White, MD
Darrell E. White

There are so many interesting things about this project, it is difficult to know where to start. To my knowledge, this is the first study to incorporate nonclinical “big data” in the evaluation of ophthalmic treatments (the Watson). In addition, the scientists from both Sun and IBM were prescient enough to realize that a “real-world” evaluation of any treatment would require the input of doctors who are engaged in that treatment (the Holmes). Paul and I were consulted before running the analysis so that appropriate endpoints were chosen. Of course, we also directed the interpretation of the results.

Every eye doctor who takes care of DED has long felt that these medicines are hard for both patients and doctors to use. We live in a “Tylenol society”: If you have a symptom, you take a medicine and the symptom goes away. Neither Restasis (cyclosporine ophthalmic emulsion 0.05%, Allergan) nor Xiidra (lifitegrast ophthalmic solution 5%, Novartis) works like that. There are frequent side effects that make Restasis and Xiidra hard to take. Both of these medicines can involve significant patient costs. In turn, these and other factors have led to poor treatment adherence. However, no one has had any real sense of just how poor that adherence truly is until now. The results of our study are eye opening.

A retrospective review utilizing a proprietary IBM database of commercial insurance claims (including Medicare patients who have secondary insurance) was performed on patients with a DED diagnosis who filled a Restasis or Xiidra prescription between July 2016 and February 2018. There was a total of 154,000 patients included, of which roughly 10,000 (6,537 for Restasis and 3,235 for Xiidra) were pure first prescriptions (eg, no prior Restasis or steroid prescriptions in the Xiidra group). This group of treatment-naive DED patients was then analyzed for treatment adherence (Did they refill their prescription?), treatment consistency (Did they switch drugs?) and treatment persistence (Were they still filling prescriptions 12 months after their first prescription?).

Paul and I made our first contributions when the group was making decisions on what would constitute adherence, persistence or discontinuation. Naturally, non-clinicians who have little insight into the mechanics of DED care would choose to define non-adherence as those patients who do not refill their first prescription at the 30-day mark. Likewise, a patient who has only obtained one refill in 12 months would be designated as a patient who has discontinued their medication. However, the reality on the ground is that every patient started on Restasis or Xiidra in the period studied received samples that covered a minimum of 4 to 6 weeks of treatment. We all know that Restasis dropperettes contain at least three doses and that the Xiidra bullets contain two. Hence, we chose 120 days for Restasis and 90 days for Xiidra as the cut-off times to declare that patients had discontinued their medication.

What did we find? Greater than 60% of the patients studied in both groups had definitively discontinued their drugs at 12 months (no refills). Taking into account drug switches (5% Restasis to Xiidra/9.6% Xiidra to Restasis), only 31.7% of Restasis patients and 27.5% of those taking Xiidra were still being treated with their originally prescribed medication. Less than a third of DED patients for whom an immunomodulator was prescribed remained under treatment at the end of 12 consecutive months.

This is a stunning outcome.

What could possibly explain this? Once again, when it came time for the discussion of the results, the real-world experience that Paul and I have in treating DED patients in the wild gave the quantitative analysts some insight. July 2016 was when Xiidra launched. We all know that a new medication in the market has minimal insurance coverage. To make matters worse, Allergan had (and continues to have) all of the Medicare coverage locked up. In the excitement of having a new DED medicine for the first time in 15 years, DED docs prescribed Xiidra for everyone, including Medicare patients. A substantial percentage of commercially insured patients (and all of the Medicare-age patients) headed for the hills when they ran afoul of the double-edged sword of no coverage and the full retail cost of Xiidra. This is certainly the main factor that explains the greater change rate from Xiidra to Restasis. Interestingly, there was a small trend toward greater adherence and persistence among Xiidra patients, although it was not statistically significant.

If Restasis was so well covered by insurance, why did so many patients dump it? As we all know, Restasis is actually not all that well covered — it was just better (and more broadly) covered than Xiidra. No amount of insurance coverage can overcome the “Tylenol society” in which we live. Neither Xiidra nor Restasis offers instant relief, and both can cause significant discomfort on instillation. And face it, the majority of doctors writing first Restasis scripts never got the memo about using adjuvant steroids to onramp Restasis. Non-adherence and non-persistence are not surprises, only the magnitude is. I think the conclusion is spot-on: We need more, more readily accessible and easier to use DED treatments in the market.

Was this an AI study? Nah. What we did was apply the new world of big data analytics to on-label medical treatment of DED. Just like Big Blue, another famous IBM supercomputer, routinely crushes international grandmasters at chess, the “real world of DED care” study utilizes “brute strength” computing to identify patterns and trends in massive data sets. It is a cool study, in part because it is first, but it is hardly the coolest stuff we are going to see if true AI ever hits our little DED world.

Frankly, I have wondered why we have not seen a study like this sooner. This type of data crunching, sometimes coupled with machine learning (see retinopathy, diabetic) has been around since the ’70s (see my upcoming blog series on big data, machine learning and AI in eye care). Wouldn’t you love to see this kind of analysis of adherence, persistence and treatment abandonment of Restasis in the pre-pharmacy benefit manager era? Why, then, has it taken so long to crunch data that has been available for decades? The power of Watson to find patterns in huge data sets, to not only find the needle in thousands of haystacks but to also tell you what direction that needle is pointing, is awe-inspiring.

Perhaps it is something that Paul Karpecki and I realized during our first meeting with our co-authors: Watson works better when you have Holmes to guide it. I hope the Watson folks do not lose my phone number.

Disclosure: White reports he is a consultant to Allergan, Shire, Sun, Kala, Ocular Science, Rendia, TearLab, Eyevance and Omeros; is a speaker for Shire, Allergan, Omeros and Sun; and has an ownership interest in Ocular Science and Eyevance.

Did you know that I am an author? Yup. Really and truly. Well, a couple of other folks helped out. Dr. Paul Karpecki and I were invited by the scientists at Sun Pharmaceutical and the IBM Watson team to help design and write a paper, “Real-world treatment patterns of cyclosporine ophthalmic emulsion and lifitegrast ophthalmic solution among patients with dry eye,” looking at dry eye disease treatment patterns. Everything about this project was super cool, especially for someone who lives in the post-drug development treatment space like me.

Darrell E. White, MD
Darrell E. White

There are so many interesting things about this project, it is difficult to know where to start. To my knowledge, this is the first study to incorporate nonclinical “big data” in the evaluation of ophthalmic treatments (the Watson). In addition, the scientists from both Sun and IBM were prescient enough to realize that a “real-world” evaluation of any treatment would require the input of doctors who are engaged in that treatment (the Holmes). Paul and I were consulted before running the analysis so that appropriate endpoints were chosen. Of course, we also directed the interpretation of the results.

Every eye doctor who takes care of DED has long felt that these medicines are hard for both patients and doctors to use. We live in a “Tylenol society”: If you have a symptom, you take a medicine and the symptom goes away. Neither Restasis (cyclosporine ophthalmic emulsion 0.05%, Allergan) nor Xiidra (lifitegrast ophthalmic solution 5%, Novartis) works like that. There are frequent side effects that make Restasis and Xiidra hard to take. Both of these medicines can involve significant patient costs. In turn, these and other factors have led to poor treatment adherence. However, no one has had any real sense of just how poor that adherence truly is until now. The results of our study are eye opening.

A retrospective review utilizing a proprietary IBM database of commercial insurance claims (including Medicare patients who have secondary insurance) was performed on patients with a DED diagnosis who filled a Restasis or Xiidra prescription between July 2016 and February 2018. There was a total of 154,000 patients included, of which roughly 10,000 (6,537 for Restasis and 3,235 for Xiidra) were pure first prescriptions (eg, no prior Restasis or steroid prescriptions in the Xiidra group). This group of treatment-naive DED patients was then analyzed for treatment adherence (Did they refill their prescription?), treatment consistency (Did they switch drugs?) and treatment persistence (Were they still filling prescriptions 12 months after their first prescription?).

PAGE BREAK

Paul and I made our first contributions when the group was making decisions on what would constitute adherence, persistence or discontinuation. Naturally, non-clinicians who have little insight into the mechanics of DED care would choose to define non-adherence as those patients who do not refill their first prescription at the 30-day mark. Likewise, a patient who has only obtained one refill in 12 months would be designated as a patient who has discontinued their medication. However, the reality on the ground is that every patient started on Restasis or Xiidra in the period studied received samples that covered a minimum of 4 to 6 weeks of treatment. We all know that Restasis dropperettes contain at least three doses and that the Xiidra bullets contain two. Hence, we chose 120 days for Restasis and 90 days for Xiidra as the cut-off times to declare that patients had discontinued their medication.

What did we find? Greater than 60% of the patients studied in both groups had definitively discontinued their drugs at 12 months (no refills). Taking into account drug switches (5% Restasis to Xiidra/9.6% Xiidra to Restasis), only 31.7% of Restasis patients and 27.5% of those taking Xiidra were still being treated with their originally prescribed medication. Less than a third of DED patients for whom an immunomodulator was prescribed remained under treatment at the end of 12 consecutive months.

This is a stunning outcome.

What could possibly explain this? Once again, when it came time for the discussion of the results, the real-world experience that Paul and I have in treating DED patients in the wild gave the quantitative analysts some insight. July 2016 was when Xiidra launched. We all know that a new medication in the market has minimal insurance coverage. To make matters worse, Allergan had (and continues to have) all of the Medicare coverage locked up. In the excitement of having a new DED medicine for the first time in 15 years, DED docs prescribed Xiidra for everyone, including Medicare patients. A substantial percentage of commercially insured patients (and all of the Medicare-age patients) headed for the hills when they ran afoul of the double-edged sword of no coverage and the full retail cost of Xiidra. This is certainly the main factor that explains the greater change rate from Xiidra to Restasis. Interestingly, there was a small trend toward greater adherence and persistence among Xiidra patients, although it was not statistically significant.

PAGE BREAK

If Restasis was so well covered by insurance, why did so many patients dump it? As we all know, Restasis is actually not all that well covered — it was just better (and more broadly) covered than Xiidra. No amount of insurance coverage can overcome the “Tylenol society” in which we live. Neither Xiidra nor Restasis offers instant relief, and both can cause significant discomfort on instillation. And face it, the majority of doctors writing first Restasis scripts never got the memo about using adjuvant steroids to onramp Restasis. Non-adherence and non-persistence are not surprises, only the magnitude is. I think the conclusion is spot-on: We need more, more readily accessible and easier to use DED treatments in the market.

Was this an AI study? Nah. What we did was apply the new world of big data analytics to on-label medical treatment of DED. Just like Big Blue, another famous IBM supercomputer, routinely crushes international grandmasters at chess, the “real world of DED care” study utilizes “brute strength” computing to identify patterns and trends in massive data sets. It is a cool study, in part because it is first, but it is hardly the coolest stuff we are going to see if true AI ever hits our little DED world.

Frankly, I have wondered why we have not seen a study like this sooner. This type of data crunching, sometimes coupled with machine learning (see retinopathy, diabetic) has been around since the ’70s (see my upcoming blog series on big data, machine learning and AI in eye care). Wouldn’t you love to see this kind of analysis of adherence, persistence and treatment abandonment of Restasis in the pre-pharmacy benefit manager era? Why, then, has it taken so long to crunch data that has been available for decades? The power of Watson to find patterns in huge data sets, to not only find the needle in thousands of haystacks but to also tell you what direction that needle is pointing, is awe-inspiring.

Perhaps it is something that Paul Karpecki and I realized during our first meeting with our co-authors: Watson works better when you have Holmes to guide it. I hope the Watson folks do not lose my phone number.

Disclosure: White reports he is a consultant to Allergan, Shire, Sun, Kala, Ocular Science, Rendia, TearLab, Eyevance and Omeros; is a speaker for Shire, Allergan, Omeros and Sun; and has an ownership interest in Ocular Science and Eyevance.