OR WAIT 15 SECS
Contrary to popular belief, physicians still can't retrieve the full medical history of every patient who walks into their offices. Nor can they zap prescriptions through a clearinghouse that shows whether patients are eligible for coverage and reviews all medications they are taking to forestall drug?drug interactions.
Contrary to popular belief, physicians still can't retrieve the full medical history of every patient who walks into their offices. Nor can they zap prescriptions through a clearinghouse that shows whether patients are eligible for coverage and reviews all medications they are taking to forestall drug–drug interactions. That era lies in the not-too-distant future. Nonetheless, pharma companies have access to vast amounts of patient data that they can use to their advantage today.
Patient data fall into two distinct groups, each serving a different market need. Patient-level data, derived from prescriptions filled and covering about half of the 3.2 billion scripts dispensed in the United States annually, can be projected to the entire US population. Patient-centric data include patients' other interactions with the healthcare establishment, such as hospitalizations and lab tests, and thus encompass more of each patient's medical history. Yet, those data represent only a sliver-in the low tens of millions-of the 200 million patients in the United States.
This article provides examples of the different types of patient data currently available and shows how pharma companies can use them to better understand the marketplace and to tap into untold pockets of opportunity. (For a list of prominent data vendors and their flagship products, see "Who's Who Among Data Vendors," page 88.)
It also describes how to select the most appropriate patient data source based on the objectives of the analysis and the therapeutic area. Finally, it shows why superior patient-level data will make patient-centric data more useful in making projections about the drug-purchasing habits of US patients.
Patient-level data are significant because they allow companies to track prescription activity in more detail than physician-level data. They are useful primarily in the areas of territory management and sales force compensation.
Physician-level data simply say, for instance, that Dr. John Smith wrote 20 prescriptions of Plavix (clopidogrel) last week. Patient-level data take the information further. First, they specify the patients' sex and age, which allows marketers to establish the demographic profile of Dr. Smith's patient base and to compare it with that of other leading Plavix prescribers.
Second, patient-level data specify the therapy the patients were on before they began taking Plavix. Marketers can then identify cases of new therapy starts (no prior therapy) and switches (patients taking a different drug before), as well as the direction of the switch (to or from), add-ons (patients adding Plavix to another therapy), and dosing changes (adjustment of medication strength).
A few weeks into a product launch, marketers armed with patient-level data can go beyond the routine task of identifying early adopters. They can establish which patients doctors are most likely to start on the new therapy, which physicians have a large number of patients that match those profiles, which drugs doctors are shunning in favor of the new launch, and when they favor an add-on to the current therapy instead of a substitution. Privy to such underlying dynamics, marketers can craft the right message for each doctor. After that, it is a question of marshalling the sales force and other promotional resources to get those targeted messages across.
Patient data can also be a potent asset in winning market share when a competitor's product is recalled. By understanding the driving forces behind the substitution, product managers can quickly develop convincing arguments to take advantage of the boon.
Bayer's recall of Baycol (cerivastatin) in 2001 offers a prime example. Competing statin manufacturers had only a small window of opportunity in which to claim the product's forfeited market share. The trophy went to the company capable of identifying Baycol users and switching them over before anyone else could flex a muscle. Merck did a great job initially, but Pfizer caught up to become the overall winner.
Patient-level data can be helpful in less volatile situations as well. Longitudinal analyses-tracking changes in a patient's medications over time-can help maintain, if not grow, market share by unveiling critical patient behavior such as switches, new therapy starts, multiple pharmacy use, and dosing, among others. Compliance and persistency are popular targets for patient-level analyses because a modest increase in either may spell huge revenue hikes for pharma companies and significant savings for insurance companies that must foot medical bills that result from non-compliance.
Companies can use patient-centric data to analyze patients' medical histories as thoroughly as possible, including their prescriptions, hospitalizations, outpatient visits, diagnoses and procedures in ambulatory facilities, lab and hospital tests, and any other information their records reveal. It is not critical to capture every single individual in a country or ensure that the data are up-to-the-minute as long as they represent the patient population to which the findings will be projected. Patient-centric data are useful for everything from disease management to strategic marketing.
By piecing together the pharmacy and medical components of patient histories, both manufacturers and payers may glean and wield big-picture insights. A pharma company might justify the high price of its drug by citing patient-centric data demonstrating that patients taking its medication are less likely to require costly hospitalization than patients taking alternative medicines, proving that the more expensive drug saves payers money. Payers could use predictive modeling-and technology similar to the neural nets credit card companies use to red-flag stolen cards-to monitor patient behavior, such as rising insulin levels, to pre-empt costly and devastating crises.
By using patient-centric data, insurance companies and pharmacy benefit managers (PBMs) can elicit best practices by establishing which therapies lead to the most favorable outcomes. Indeed, they have access to countless cases in which different therapies are administered to patients that share similar medical situations and, more important, to the results that ensue. Shifting the focus to healthcare providers, payers can grade the former based on their use or non-use of those best-practice therapies.
Another promising area is the identification of untreated patients, an untapped market that companies can access immediately.
Patient data come in different forms because they are captured at different points: at the pharmacy/switch-the computer system that connects pharmacies together and supports the adjudication process that ensues when patients tender their PBM cards-by PBMs, by payers, or by healthcare providers. A piece of information gathered at one point is likely to be missed altogether at another. That results in data source differences, such as the number of patients or scripts recorded, timeliness of the data, and geographic coverage. Those differences create competition among data vendors and confusion for data buyers new to the market.
Pharmacies can record cash payments for a product, but payers don't capture that information because someone else foots the bill. Yet payers know patient eligibility well, because it is the basis for accepting or declining payments, whereas the pharmacy/switch is unconcerned with where the money comes from.
Data vendors ply their wares by stressing the type of information they can provide, particularly patient eligibility, cash payments, patient-centricity, and connectivity. Understanding those features and their ramifications is important, because they have a direct bearing on the quality of the analyses. Therefore, it behooves analysts to ensure that the most appropriate database is deployed to answer the sales and marketing questions posed. (See "Source Checklist.")
Eligibility. According to AdvancePCS, eligibility data are essential for trustworthy persistency analyses. Without such data, it is nearly impossible to ascertain whether patients drop a therapy for medical reasons or simply because they no longer have insurance coverage. For that reason, failing to track the eligibility of all patients under study from start to finish may strongly skew the findings, because they may not represent the broader patient population, just as data about people who eat in five-star restaurants are unlikely to shed light on the eating habits of the average Joe.
Nevertheless, data vendors may infer patient eligibility by tracking families rather than individuals. If a family member has a prescription filled and reimbursed, chances are that other family members have insurance as well. Although such inferences have their limitations, they can provide insight into many cases.
Cash payments. Cash payments are more important than they may seem, according to those who capture the data at the pharmacy/switch level. Leaving them out punches holes in the longitudinal data and sometimes contaminates analyses altogether. Patients pay cash for prescription medications for many reasons, including:
Patient-centricity. This type of patient data centers on individual patients and emphasizes completeness by capturing not only drug prescriptions but also medical expenses, hospitalizations, and lab tests. For many therapeutic areas, such as cardiovascular, tracking prescriptions is only part of the equation, which can also include such factors as lab tests, hospitalizations, and clinic visits.
Payer and employer group data are certainly a very good source of patient data because they track every penny, whether pharmaceutical or medical, expended on each patient. Some argue that healthcare provider data are even better. If a patient is admitted to the hospital, diagnosed with several medical conditions, treated for one and discharged, the payer's claim data will pick up only the code that corresponds to the patient's release, because that is the only information needed to process the payment. Yet the hospital's data will capture all the other diagnoses as well. For that reason, provider data may sometimes be a better source, as in the case of co-morbidity analyses.
Connectivity. This feature refers to the fact that patients' actual medical encounters with healthcare providers-physicians, pharmacists, lab technicians, nurses, hospital personnel-are captured in the database and actually recorded as pertaining to that patient. That is far from being obvious. Unless special procedures are put in place and dedicated efforts are expended to achieve that goal, the tracking system may not recognize that one transaction in one part of the system-say, the prescription part-refers to the same entity in another part of the system, such as the medical part.
That is not a sure thing even for patient-level data, even though it focuses entirely on drug prescriptions. It is not easy to be consistent in the tracking process, especially when the patient's spouse picks up the medication after the patient selects a pharmacy from a different chain-Walgreen's, say, instead of Albertson's-or when the dispensing pharmacy is located in another part of the country. In fact, connectivity is as important, if not more important, than metrics such as the number of covered patients or good representation in all geographic areas.
Taking connectivity one step further, PharMetrics offers a broad view of the patients it tracks by covering not only pharmaceutical records but also medical histories, including hospitalizations, lab tests, and visits to ambulatory centers.
With records of 65 million patients, Medco-through its wholly owned subsidiary TIER-claims to have the largest database, followed by AdvancePCS, whose 75 million patient records come from two sources, Advance Paradigm and PCS. PharMetrics claims to track 27 million patients as a result of its agreements with 54 or so plans. They do not disclose the identity of those plans, prompting detractors to raise concerns regarding geographic bias and PharMetric's ability to seamlessly integrate data from disparate sources.
Ingenix, the wholly owned subsidiary of United Healthcare, on the other hand, does not face such an issue, because its 44 constituent plans all come from the same company. With only ten million patients on record, however, it is relatively small and may be geographically skewed.
Recognizing the therapies a patient takes is a no-brainer for payers, but from the pharmacy/switch vantage point, transactions are more complex. If a patient who is currently taking asthma medication uses his card to pick up a refill for his diabetic wife, for instance, some major misrepresentation could contaminate the data unless special care is taken. Not only could the diabetes medication be wrongly assigned to the asthma patient, but his wife also could be proclaimed non-persistent because she failed to fill her own diabetes script.
Currently, there is little difference between the types of analyses that data vendors say their data enable-they're all longitudinal. All the vendors promote the same list of offerings, including such mainstays as new therapy starts, compliance, persistency, and product switches. Yet patient-level data can do much more.
They can, for instance, identify influencers, their spheres of influence, and their referral patterns. The best physicians to target may be not only high prescribers but also those who influence them. The bulk of many high prescribers' scripts are for refills of prescriptions initiated by other doctors. Therefore, detailing them may not be the best use of reps' time. On the other hand, some physicians-opinion leaders, hospital heads, group practice leaders, and lead investigators of clinical trials-may write few scripts but have a great deal of influence on other physicians. Companies often assign a dedicated sales force to win over those important physicians with science-heavy dialogue.
Primary research can, to a large extent, identify such medical thought leaders. In essence, it involves talking with a very large number of professionals and enlisting the probing capabilities of the field force. But patient-level data open up a host of avenues for tracking down those influencers.
A patient can be a link between two physicians. By tracking the flow of patients from one doctor to another, data analysts may elicit insights about relationships among physicians: who sends patients to whom, who belongs to the high-profile inner circles, who initiates therapies that patients tend to renew repeatedly even when they bounce from provider to provider. By bringing in additional information such as group practice membership, managed care affiliation, and composition of lead investigational teams, analysts can, with great precision, identify those influencers and how they shape trends in the physician community. In that way, mining patient-level data may uncover yet untold ways to better approach key influencers and make inroads into murky areas such as off-label usage.
A 1999 Institute of Medicine study revealed that gross inefficiencies resulting from the use of pen and paper to record medical information lead to 1.3 million injuries and 7,000 deaths a year, and cost healthcare providers and insurers $77 billion. Last year alone, pharmacists placed 150 million follow-up phone calls to physicians to clarify prescriptions. That contributed to administrative inefficiencies adding up to 15–25 percent of the annual $1.2 trillion US healthcare expenditure.
In February 2001, those shortcom-ings prompted the three largest PBMs-AdvancePCS, Medco Health, and Express Scripts-to join forces. The result was RxHub, an online data clearinghouse that links physicians' offices to pharmacies. RxHub's three PBM partners represent a whopping 185 million consumers. So, even if physicians are slow to embrace electronic prescriptions-less than 10 percent of scripts are digital-big companies are demand-ing that patient data go high-tech, speeding its capture and rendering it more accurate in the process.
Yet, the path to that goal is not free of obstacles. Pharmacists are concerned that RxHub will divert scripts intended for brick-and-mortar pharmacies to its member PBMs' mail-order facilities. In response to that threat, the National Community Pharmacists Association (NCPA) and the National Association of Chain Drug Stores (NACDS) founded SureScript in August 2001 to focus on refills, which make up 80 percent of all transactions. NCPA alone represents nearly 25,000 independent community pharmacies, accounting for nearly $50 billion in prescription sales.
The claims processing industry is undergoing changes as well. Several major health plans-including Aetna, Anthem, CIGNA, Health Net, Oxford, PacifiCare, and WellPoint Health Networks-came together to form MedUnite, a conglomerate with a simple objective: to take on WebMD, the lone middleman standing between them and physicians. Nevertheless, MedUnite is not yet a good source of patient-centric data, because it currently processes only a small portion of its founders' claims.
Other undertakings aim to connect multiple healthcare providers in a given geographic area, thereby producing high-quality patient-centric data. One such program is CareScience, which uses peer-to-peer technology-in which one computer links directly to another without going through an intermediary server-to connect medical groups, hospitals, clinics, and other healthcare organizations. As a result, 75 percent of the healthcare providers in Santa Barbara, California, have access to information about 300,000 patients.
The dilemma of patient data is that it must be both timely and comprehensive, and that is a paradox. In pursuit of broader coverage, data providers will have to partner with a host of other data providers. And, because the smaller partners have less capital-intensive infrastructures, they are bound to operate at lower speeds, reducing the overall efficiency. Moreover, those partners may use different data formats and protocols, forcing companies to waste time cleaning and standardizing the data.
That will force vendors to shift their positioning to coverage and redefine timeliness in terms of days rather than minutes. Manufacturers will base their go/no-go purchase decisions on coverage, and the litmus test will be whether the data-after being chopped into 500 or so territories-still accurately capture patient activity.
Yet coverage will present a big challenge to data vendors for some time to come. Consider the type of prescription activity that IMS Health and NDC Health collect: "Dr. So-and-So wrote 16.73-not 16, not 17-scripts of Zocor (simvastatin) last week." Those data are obviously projected. And, considering that the number of US prescribers is 300 times smaller than the potential national patient population, it's no surprise that the patient-level data market is so complex.
Despite having incomplete raw data, all data vendors must offer the most comprehensive data possible to customers. That means that the vendor wielding the best projection methodology will have a formidable competitive advantage.
An approach that incorporates both patient-centric and patient-level data should succeed, because it will yield insightful projections that analysts can apply to the whole patient data universe. In turn, they can extrapolate desirable patient treatment patterns and outcomes observed in the patient-centric data to the larger but less comprehensive patient-level data.
In the same vein, physician-level data products such as IMS Xponent or NDC Source Prescriber can use the profiles of specific physicians-their specialties, managed care and hospital affiliations, association memberships, and interests-as a template to identify a full range of physician behavior. Because that kind of analysis requires patient-centric information, the industry will see patient-level vendors ink partnership deals with patient-centric players to gain access to their data.
Compliance with the Health Insurance Portability and Accountability Act (HIPAA) is a top priority. Companies are very uncomfortable housing, let alone handling, patient data whenever there is a chance that someone may be able to identify a patient on file either directly or through a third-party database. The privacy rules that went into effect on April 14 this year also require pharmacies, health plans, and other covered entities to obtain specific authorization from patients before sending them promotional materials.
Those privacy rules will have two short-term effects:
1. Companies unsure of their public image and worried about being perceived as bending privacy rules will have to wait to see if there is a sufficient market to warrant the risk. Among the companies that have decided not to jump in yet is Express Scripts, the third largest PBM in the United States, with 45 million enrollees. But data providers determined to stay the course understand that HIPAA and bad press are simply the cost of doing business.
2. Companies will shift their offerings from bulky databases of detailed patient records to statistics on aggregate patient cohorts. At the same time, to sidestep the need to hand them over to customers, they will emphasize one-stop shopping and underscore the value of analytics-a field still in its infancy-as opposed to the simple data. Medco Health sells only aggregate patient-level data as off-the-shelf reports and is positioning itself as a consultancy through its TIER subsidiary. However, that premature shift of emphasis from data to analysis will rob pharma companies-and the third parties working on their behalf-of the opportunity to peruse the data, pinpoint weaknesses and blind spots, and establish whether the analyses' end points have been compromised. If the data supplier performs the analysis as well, it will be tempted to obscure such shortcomings.
An even worse foe than HIPAA is the healthcare industry's continuing reliance on pen and paper. Ironically, the paper system's tendency to cause inefficiencies and even fatal inaccuracies is a major rationale for digitizing all healthcare records and prescriptions.
Another difficulty facing the market is data-rich pharma companies' reluctance to pay for even more information. Unless they see some dramatic examples of what patient data have to deliver, they may dismiss the whole idea as a passing fad. (See "Success Stories," page 90.)
But make no mistake: A great future awaits patient-level data. In the early 1990s, the industry switched to physician-level prescription data to make up for the shortcomings of outlet-level data, and it is now poised to make a similar move to patient-level data. But this time, the implications for sales and marketing analyses will be even greater. Patient data will have finally come of age.