Using heuristics to identify drivers of perceived value Laurie Gelb
Early-stage market research needs <ul><li>A systematic, robust method to forecast value for a compound given information a...
Besides the forecast, why do early-stage research?  <ul><li>To understand threshold values and intangibles that drive curr...
Drug decisions: unusually difficult to model <ul><li>Specifiers (HCPs), purchasers (payors) and consumers (patients) diffe...
Drugs defy attribute “deconstruction” <ul><li>Neither pharma manufacturers nor consumers can easily implement changes in a...
Whereas other goods easily permit deconstruction <ul><li>It’s easy to “build your own” car or even house with new search e...
HCPs treat pts, not dz  (& implement tx plans, not just drugs) <ul><li>Decades ago, Dr. William Osler noted that many seem...
Nor is there a “typical” physician <ul><li>If we accept that “the normal is the ideal,” we shouldn’t waste time chasing th...
Desired rx outcomes: how relevant are aggregate data? <ul><li>Physician wants problem resolved clinically and to patients’...
Rx outcome measures in practice  <ul><li>“ No news is good news” </li></ul><ul><ul><li>Patient satisfaction is presumed, s...
Rx/tx decisions:  always made in ignorance <ul><li>Whereas there is little more to know about Raisin Bran ®  after a singl...
<ul><li>Heuristics say no: we cut to the chase, and a busy physician or a frazzled patient poised to flee an exam room mos...
When conjoint works <ul><li>Conjoint is applicable when attributes are intrinsic to the product, and when the mfr  cannot ...
Static profile drawbacks <ul><li>Several important decision drivers, e.g. formulary status, cannot be generalized </li></u...
What’s more important than varying attribute levels?  <ul><li>We already know whether we have a product from the historica...
Heuristics are only human <ul><li>Every human being uses “heuristics” (shortcuts). These are subject to: </li></ul><ul><ul...
Heuristics in real life <ul><li>Reference values: you look for soups that contain less than 1000mg of Na per serving or a ...
Heuristic rx inputs <ul><li>Physician makes provisional evaluation; settles on class of drug(s) to rx, then pressure and c...
Typical outputs from static profile studies <ul><li>Many reactions to each preset level, with no basis for interpolation o...
<ul><li>Efficacy vs. safety </li></ul><ul><li>Formulation vs. efficacy  </li></ul><ul><li>Cost vs. efficacy </li></ul><ul>...
Perils of “importance utilities” <ul><li>Asking for importance in the aggregate belies the uniqueness of pts, e.g. HPA axi...
Side effect incidence especially problematic <ul><li>Is it the often-used “incidence of headache” that constrains rx, or t...
Depending on system, rx-to-pt match is unpredictable <ul><li>Public payors may restrict rx to labeled indications </li></u...
The heuristic approach <ul><li>Assume that heuristics, not reasoned or conscious simultaneous tradeoffs, rule </li></ul><u...
Threshold measures <ul><li>Traditional conjoint provides for each respondent: </li></ul><ul><ul><li>Profile 1 preference s...
So, how do we value better side effect profiles?  <ul><li>Without quantifying them to the n th  degree in terms of frequen...
Need we model viability from tolerability? <ul><li>Often, regulatory considerations and legal exposure are “no-go” trigger...
What about multiple endpoints? <ul><li>Using the respondent’s salient domains, measures and thresholds can elicit more tha...
What about multiple endpoints?  (cont’d) <ul><li>Later, when you know more about your drug, you can adjust the total choic...
The power of “what else” <ul><li>Every study needs the ability to eliminate domains from consideration </li></ul><ul><li>A...
Often, context=absolute importance <ul><li>Many “attributes” should be dealt with on a “real life” basis </li></ul><ul><ul...
Crucial questions that often  aren’t  asked <ul><li>% of relevant pts for whom co-pay differential or lack of rx coverage ...
Do more frequent, less costly studies to evolve the models <ul><li>“ Not to decide is to decide” </li></ul><ul><ul><li>Ask...
“ Ping” the market more often <ul><li>Heuristic studies are quicker and cheaper than conjoint [more on this in other prese...
Summary <ul><li>Just as computer-administered surveys have replaced card sorts, it’s time for respondent-driven questions ...
Upcoming SlideShare
Loading in …5
×

Quantifying Value Drivers for Biopharmaceutical Products

1,149 views

Published on

An overview of the use of heuristic market research methods, instead of conjoint analysis, as new product forecast inputs, with a focus on biopharmaceutical applications

Published in: Business, Health & Medicine
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
1,149
On SlideShare
0
From Embeds
0
Number of Embeds
15
Actions
Shares
0
Downloads
0
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

Quantifying Value Drivers for Biopharmaceutical Products

  1. 1. Using heuristics to identify drivers of perceived value Laurie Gelb
  2. 2. Early-stage market research needs <ul><li>A systematic, robust method to forecast value for a compound given information available at this time </li></ul><ul><ul><li>Knowing many of our assumptions will likely unravel later </li></ul></ul><ul><ul><li>But knowing also that early forecasts will drive critical decisions, including go/no-go </li></ul></ul><ul><li>An evidence-based means of guiding early clinical development and market strategy </li></ul>
  3. 3. Besides the forecast, why do early-stage research? <ul><li>To understand threshold values and intangibles that drive current decisions, and how these may change </li></ul><ul><ul><li>As well as likely interactions among factors driving reimbursement, prescribing and care-seeking </li></ul></ul><ul><li>To learn about drivers of action or preference share, and the factors other than our product’s attributes that will drive behavior </li></ul>
  4. 4. Drug decisions: unusually difficult to model <ul><li>Specifiers (HCPs), purchasers (payors) and consumers (patients) differ most of the time </li></ul><ul><li>High stakes: wrong decisions can impair, disable or kill </li></ul><ul><li>Representational data: “selling points” = someone else’s outcomes in legal language, whereas product benefits cannot be shared </li></ul><ul><li>Ongoing re-evaluation of decisions possible based on stimuli and experiences; can stop taking rx without notice </li></ul>
  5. 5. Drugs defy attribute “deconstruction” <ul><li>Neither pharma manufacturers nor consumers can easily implement changes in attributes or levels </li></ul><ul><ul><li>Dosage form and tablet color may be the only potential “options” for commercializing a molecule </li></ul></ul><ul><li>Yet attribute levels are often interdependent, e.g. greater efficacy often implies potential toxicity </li></ul>
  6. 6. Whereas other goods easily permit deconstruction <ul><li>It’s easy to “build your own” car or even house with new search engines </li></ul><ul><li>And each attribute and level is a necessary choice: You must choose to add the anti-rust coating or not, or whether a porch is a must </li></ul><ul><ul><li>Yet if you later find the porch criterion is too limiting, you can remove it from your algorithm without affecting any other options </li></ul></ul>
  7. 7. HCPs treat pts, not dz (& implement tx plans, not just drugs) <ul><li>Decades ago, Dr. William Osler noted that many seemingly contradictory attributes are often found in the same person, and that the “normal” is an ideal, not reality. He concluded: </li></ul><ul><ul><li>“ It is much more important to know what sort of patient has a disease, than what sort of disease a patient has.” </li></ul></ul><ul><li>Osler also noted that rarely does one see a patient without “abnormal” results or findings </li></ul>
  8. 8. Nor is there a “typical” physician <ul><li>If we accept that “the normal is the ideal,” we shouldn’t waste time chasing the non-existent “typical” physician, but rather describing and influencing the full spectrum of beliefs and behavior across physicians </li></ul><ul><li>Raising the urgency of relating preference share and other measures of potential value to other survey variables, e.g. patient mix, current tx protocols, heuristics used </li></ul>
  9. 9. Desired rx outcomes: how relevant are aggregate data? <ul><li>Physician wants problem resolved clinically and to patients’ satisfaction to the maximum extent possible </li></ul><ul><ul><li>With minimal callbacks, side effects and complications </li></ul></ul><ul><li>Patients want problem resolved </li></ul><ul><ul><li>To feel better, less worried (about both problems and treatment) with minimal negative short or long-term consequences </li></ul></ul>
  10. 10. Rx outcome measures in practice <ul><li>“ No news is good news” </li></ul><ul><ul><li>Patient satisfaction is presumed, seldom verified; outcomes judged by lab values, physical exam, events, not clinical trial-style aggregate results </li></ul></ul><ul><li>“ As good as it gets” </li></ul><ul><ul><li>Patients and HCPs adjust expectations for efficacy, tolerability based on experience, external information and stimuli (friend says, “I started taking Ambien ® and I sleep for six hours now!”) </li></ul></ul>
  11. 11. Rx/tx decisions: always made in ignorance <ul><li>Whereas there is little more to know about Raisin Bran ® after a single trial, a physician’s ability to predict future individual drug experiences is severely limited, nor can pts predict their own responses much better </li></ul><ul><ul><li>Tolerance, rebound, unforeseen AEs, interactions limit predictability </li></ul></ul><ul><li>Thus, decisions are based on predictions, and we must calibrate respondents’ preferences to their initial assumptions </li></ul>
  12. 12. <ul><li>Heuristics say no: we cut to the chase, and a busy physician or a frazzled patient poised to flee an exam room most of all </li></ul><ul><ul><li>Without full arrays of rx, we see drivers (e.g. samples, pt preference, contraindication) considered sequentially— in human brain-size “bites” </li></ul></ul><ul><ul><li>Thresholds determine the set considered as well as prioritization (“Given an LDL this high, I can’t use Lescol ® , because…”) </li></ul></ul>Are full-profile tradeoffs the best decision models?
  13. 13. When conjoint works <ul><li>Conjoint is applicable when attributes are intrinsic to the product, and when the mfr cannot address them all but could potentially address any of them </li></ul><ul><ul><li>For example, conjoint has successfully designed oilfield tools and opera subscription series </li></ul></ul><ul><ul><li>Because the attributes are clear, and their presence/absence can be unequivocally defined as well as incorporated into product design </li></ul></ul>
  14. 14. Static profile drawbacks <ul><li>Several important decision drivers, e.g. formulary status, cannot be generalized </li></ul><ul><li>Inability to include all salient attributes due to respondent overload </li></ul><ul><li>Lack of evidence base for levels to include, particularly for intermediate ones </li></ul><ul><li>Must assume that consideration of all attributes is simultaneous, knowing that at least some are considered sequentially (e.g. contraindications before efficacy) </li></ul>
  15. 15. What’s more important than varying attribute levels? <ul><li>We already know whether we have a product from the historical standpoint: what our and other trials have shown </li></ul><ul><li>But we don’t know whether the market knows; studies show targets often unaware of established benchmarks </li></ul><ul><ul><li>At what point(s) does the market know we have a product, and at what point(s) must we try to educate/shape the market? </li></ul></ul><ul><ul><li>In what order will attributes be considered? </li></ul></ul>
  16. 16. Heuristics are only human <ul><li>Every human being uses “heuristics” (shortcuts). These are subject to: </li></ul><ul><ul><li>Anchoring and adjustment—initial exposures set expectations, e.g. if initial patients on Actos do better than those on Avandia, their physician is more tolerant of a future Actos patient’s failure </li></ul></ul><ul><ul><li>Priming effects—something “primes” you to interpret later events differently, e.g. after a friend’s practical joke, you suspect all strange phone calls </li></ul></ul>
  17. 17. Heuristics in real life <ul><li>Reference values: you look for soups that contain less than 1000mg of Na per serving or a car whose total sticker price is <[x] </li></ul><ul><li>Representativeness: observations of one group are taken to apply to another; a physician makes a presumptive dx given a cluster of sx, based on previous cases </li></ul><ul><li>Proportionality—you see only part of a phenomenon, and therefore misattribute its cause (chickens and eggs) </li></ul>
  18. 18. Heuristic rx inputs <ul><li>Physician makes provisional evaluation; settles on class of drug(s) to rx, then pressure and constraints come into play: </li></ul><ul><ul><li>Patient fits the preferred or contraindicated population for a drug </li></ul></ul><ul><ul><li>Patient suggests or asks about drug </li></ul></ul><ul><ul><li>Payor prefers and/or constrains agent(s) </li></ul></ul><ul><ul><li>One drug is indicated and another not </li></ul></ul><ul><ul><li>Supply of samples </li></ul></ul><ul><ul><li>Some “spread around” rx’ing within a class </li></ul></ul>
  19. 19. Typical outputs from static profile studies <ul><li>Many reactions to each preset level, with no basis for interpolation of other levels, within a range that may be far longer (or shorter) than most respondents would use </li></ul><ul><li>No information on reference points or threshold values (how did our 36% reduction in exacerbations compare with what physicians think they see with existing agents?) </li></ul>
  20. 20. <ul><li>Efficacy vs. safety </li></ul><ul><li>Formulation vs. efficacy </li></ul><ul><li>Cost vs. efficacy </li></ul><ul><li>New brand vs. old generic </li></ul><ul><li>Mono vs. combo therapy </li></ul><ul><li>“ Watch & wait” vs. “treat early” </li></ul>Pharma’s false dichotomies
  21. 21. Perils of “importance utilities” <ul><li>Asking for importance in the aggregate belies the uniqueness of pts, e.g. HPA axis suppression is most relevant in children, drug interactions relate to polypharm, safety in elderly most important for… </li></ul><ul><li>So what you are really getting from conjoint is an amalgamation of patient mix and preferences, with no way to separate the two </li></ul>
  22. 22. Side effect incidence especially problematic <ul><li>Is it the often-used “incidence of headache” that constrains rx, or the severity, frequency and duration of headache, e.g. whether or not a patient complains or discontinues the drug? </li></ul><ul><ul><li>It’s the patient who decides that a headache is “severe” enough to report and/or do something about </li></ul></ul>
  23. 23. Depending on system, rx-to-pt match is unpredictable <ul><li>Public payors may restrict rx to labeled indications </li></ul><ul><li>Length, quantity of therapy constraints inhibit free exercise of clinical judgment </li></ul><ul><li>Patient access to physicians and tests is also limited; can be long delay from presentation of symptoms to visit(s) for dx/tx </li></ul><ul><ul><li>Thus, regimens requiring intensive follow-up, e.g. titration, are less often possible, physicians’ preferences notwithstanding </li></ul></ul>
  24. 24. The heuristic approach <ul><li>Assume that heuristics, not reasoned or conscious simultaneous tradeoffs, rule </li></ul><ul><li>Elicit heuristics, reference values for heuristics (completely unaided, not as multiple choices) and how they’re used </li></ul><ul><li>Respondents validate the attributes that matter to them, report their own thresholds and express their interest in a drug or treatment on that basis </li></ul>
  25. 25. Threshold measures <ul><li>Traditional conjoint provides for each respondent: </li></ul><ul><ul><li>Profile 1 preference share = XX% </li></ul></ul><ul><ul><li>Profile 2 preference share = YY% </li></ul></ul><ul><li>Heuristic studies provide thresholds (minimum levels) for each respondent: </li></ul><ul><ul><li>Profile preference share with respondent’s thresholds = XX% [TPP can also be presented if significantly different] </li></ul></ul><ul><ul><li>Attribute 1 threshold (if applicable) = Z </li></ul></ul><ul><ul><li>Attribute 2 threshold (if applicable) = C </li></ul></ul>
  26. 26. So, how do we value better side effect profiles? <ul><li>Without quantifying them to the n th degree in terms of frequency, duration, intensity, effect on function, we can’t </li></ul><ul><ul><li>It’s like asking a computer owner to value different error messages – what happens after the error message is what’s important </li></ul></ul>
  27. 27. Need we model viability from tolerability? <ul><li>Often, regulatory considerations and legal exposure are “no-go” triggers </li></ul><ul><ul><li>Do we have trouble deciding whether to proceed with development at a certain headache level, assuming efficacy has a demonstrated value? </li></ul></ul><ul><li>We do not need respondents to set our decision rules here but careful study of the clinical literature, promotional/regulatory consequences and common sense </li></ul><ul><ul><li>The raison d’etre of market research is actionability </li></ul></ul>
  28. 28. What about multiple endpoints? <ul><li>Using the respondent’s salient domains, measures and thresholds can elicit more than preference share assigned to a custom profile </li></ul><ul><ul><li>Relative importance within the profile can also be assessed </li></ul></ul><ul><li>Pre-profile tasks also help assess “necessary” vs. “sufficient” vs. “nice” attributes; patient mix may de facto weight shares </li></ul>
  29. 29. What about multiple endpoints? (cont’d) <ul><li>Later, when you know more about your drug, you can adjust the total choice share/eligible patients for your forecast </li></ul><ul><li>Drawing only from respondents for whom the domains, measures and thresholds of the new “real” (approvable) profile were salient </li></ul>
  30. 30. The power of “what else” <ul><li>Every study needs the ability to eliminate domains from consideration </li></ul><ul><li>As well as the ability to specify other measures for each attribute, and appropriate levels for one’s own measure or the presented one(s) </li></ul><ul><ul><li>Are you starting with pain, swelling or something else? </li></ul></ul><ul><ul><li>How are you measuring your key drivers? </li></ul></ul><ul><ul><li>What’s your reference point for each measure? </li></ul></ul>
  31. 31. Often, context=absolute importance <ul><li>Many “attributes” should be dealt with on a “real life” basis </li></ul><ul><ul><li>Quantify the present: what % of Zyprexa pts are on the ODT formulation, what % of pts on antipsychotics are noncompliant on QD formulations, what % of asthma pts are 4-12, what % of MS pts are female, what % are still considering pregnancy, etc. </li></ul></ul><ul><li>Reality today is an important predictor of value tomorrow! </li></ul>
  32. 32. Crucial questions that often aren’t asked <ul><li>% of relevant pts for whom co-pay differential or lack of rx coverage is issue </li></ul><ul><li>% of time physician knows/does not know reimbursement status before prescribing </li></ul><ul><li>% of time pt requests drug, form or dosing… (could be: I work for BMS…hate needles…am sensitive to light) </li></ul><ul><li>What drives whether pts are offered option(s) or not </li></ul>
  33. 33. Do more frequent, less costly studies to evolve the models <ul><li>“ Not to decide is to decide” </li></ul><ul><ul><li>Ask about gradual and recent changes in dx/rx/tx or lack thereof, and explicitly about expectations vs. reality of new products </li></ul></ul><ul><li>Include demographic, attitudinal and behavioral variables to facilitate both a priori and post hoc segmentation (re-evaluate schemes as we go) </li></ul>
  34. 34. “ Ping” the market more often <ul><li>Heuristic studies are quicker and cheaper than conjoint [more on this in other presentations] </li></ul><ul><ul><li>With no “black boxes” </li></ul></ul><ul><li>So you can afford to do more studies, among more audiences (interviewed any oncology PAs or GYN nurse-practitioners lately?) </li></ul><ul><li>Heuristics can also be applied to sufferers, payors and influencers, though this presentation has focused on physicians </li></ul>
  35. 35. Summary <ul><li>Just as computer-administered surveys have replaced card sorts, it’s time for respondent-driven questions to drive mfr decisions </li></ul><ul><ul><li>Static profiles will ipso facto fail to capture the domains, measures and threshold values that are salient for a respondent group, no matter how many choices within response sets are provided </li></ul></ul><ul><li>As the rest of R&D advances, if market research cannot follow suit, it deserves to lose its seat at the table </li></ul>

×