Demand Trend Improves Starting 4Q, Ortho & Commodity Suppliers Benefit Most; Why HDL-C Drugs May Be Bigger than You Think
Richard Evans / Scott Hinds
richard@ / firstname.lastname@example.org
December 1, 2010
Demand Trend Improves Starting 4Q, Ortho & Commodity Suppliers Benefit Most; Why HDL-C Drugs May Be Bigger than You Think
- High rates of job loss drive employees to pull healthcare demand forward; the effect continues for as many as 3 months after jobs stabilize. 2Q09 marked the peak rate of job losses, and excess healthcare demand likely extended into 3Q09 – thus presumably 3Q10 compared against the ‘last worst’ comparable. Comparisons ease substantially from here; by 2Q11 job-loss effects should be largely out of year-prior comparables
- Discretionary consumption is the most likely to be pulled forward; we show that orthopedic and neurologic (particularly pain-related) diagnoses are most affected, which argues ortho results have suffered most from inflated comps
- We reiterate that the impact of job-losses on current utilization trends appears at least 8 times larger than the impact of consumers moving to less generous forms of insurance; and, that the job-loss effect is short-lived
- Yet stocks price-in slow utilization as a longer-term reality; thru-out the period of difficult comps out-year (thru- ’13) revenue and EBIT estimates have fallen for utilization-sensitive sub-sectors, as have P/E forwards on these years’ estimates
- We favor a basket of commodity-oriented suppliers (e.g. BAX, BDX, BCR, COV, HSP and others) primarily, and ortho names (e.g. ZMH, SNN, SYK, NUVA) secondarily. Both groups have been similarly re-rated; ortho volumes should improve more markedly than commodity-oriented suppliers’, though ortho faces another possible round of export market pricing pressures
- The most common target for LDL-C (i.e. bad) cholesterol is 100 mg/dl, roughly 20 percent below the population average of 123 mg/dl. 43 percent of US adults meet this (and other) criteria for LDL-C lowering drug therapy; the result being a $20B (at peak) US market, and a $35B global market
- The population average for HDL-C (i.e. good) cholesterol is 51 mg/dl, assuming a target of 60 mg/dl (roughly 20 percent above the pop’n mean), 76 percent of US adults would need drug therapy to raise their HDL-C values – almost twice (1.8x) the number that need statins to lower LDL-C. Even at a very modest 40 mg/dl target, unit demand would be roughly 1/2 current statin demand
- MRK’s anacetrapib leads a field of roughly 10 known compounds in this class (CETP-inhibitors); anacetrapib is one of only 2 compounds to have reached phase III testing, and looks to be roughly 4x more potent than the other phase III compound (Roche/JT’s dalcetrapib)
As we’ve argued for some time, the present slow-down in US healthcare demand ties to changes in employment — specifically, when the economy is rapidly losing jobs, employees with employer-sponsored health insurance fear losing that coverage, and so tend to pull discretionary demand forward. We’ve shown solid statistical evidence of this effect during the ’01 – ’03 downturn, and have argued that the more recent period of job loss extending from roughly mid-’08 to mid-’09 also should have been characterized by accelerated discretionary demand. As a result, we believe that healthcare companies’ reported growth rates in the last several quarters almost certainly have been burdened by inflated comparables.
In the interest of more actionable investment conclusions, since finding this relationship between job loss and healthcare consumption, we’ve been working to refine our understanding of two things: 1) timing of the relationship between job loss and healthcare demand, i.e. whether changes in jobs and changes in healthcare demand are concurrent or lagged; and, 2) the relative importance of job loss effects on healthcare demand by diagnosis (and thus by healthcare sub-sector).
As Regards Timing ...
We looked at timing from two perspectives, individual and macro. For the individual view, we looked episodes of care among individuals who (1) had at least one encounter with a health professional; and (2) lost private insurance coverage during a calendar year. Focusing on the 2001 – 2003 period of job loss, 42 percent of episodes of care occurred during the final two months of coverage (which on average only represented 28 percent of the total insured period). Thus individual patterns of job loss and healthcare consumption are consistent with the general thesis, and suggest a comparatively tight (+/- 2 month) timing relationship.
Turning to the macro view, and using monthly observations from the ’01 – ’03 downturn, we compared changing rates of job loss nationally to changing rates of ortho / arthritis demand nationally, using various lags between jobs and demand. In effect, we’re simply stretching the lag to a point at which the delta-jobs / delta-healthcare relationship no longer exists – which turns out to be 4 months (Exhibit 1). All in, the macro findings argue that some lag is likely, but that the lag is 3 months or less – which is generally consistent with the analysis at the individual level. Thus it’s very likely that in the more recent period of job-loss (3Q’08 – 2Q’09), employees would still have been pulling healthcare consumption forward into 3Q’09. Accordingly, 3Q’10 reported growth rates are likely to have been lowered by job-loss related demand in the comparable period, but 4Q’10 results should not be as affected by an inflated comparable, and by 2Q11 job-loss effects should be largely out of year-prior comparables.
Which Diagnoses (and Sub-Sectors) are More or Less Affected by Job-Loss?
Clearly any demand that is consciously pulled forward for fear of losing insurance at least begins as discretionary demand (e.g. arthritis), though we note that an increased rate of patient – physician interaction also must lead to higher rates of less discretionary (e.g. heart disease) or even non-discretionary ‘events’ (e.g. cancer), simply by virtue of physicians having greater opportunity to diagnose conditions for which care is more urgently needed.
We looked for relationships between job loss and healthcare demand, by diagnosis and/or sub-sector, using both Medical Expenditure Panel Survey (MEPS) data and company-reported sales.
The MEPS View
MEPS data offer far more observations and detail than do company reported sales, but do not yet cover the most recent downturn. Using MEPS event data, we estimated the number of episodes of care per diagnosis per month during the 2001 – 2003 employment down-cycle. The estimated total number of events is illustrated in Exhibit 2 for several categories. In addition, we looked at the net y/y change in total full-time employment (also by month) over the same period (Exhibit 3). We then regressed changes in diagnosis frequency with changes in employment. Controlling for variations in seasonality (e.g., respiratory events increase during the winter; cardiology and neurology do not), and autocorrelation, results from the 2001 – 2003 period are presented in Exhibit 4. By way of orientation, note that a statistically significant negative coefficient is consistent with our hypothesis of a pull forward (the expression of the explanatory variable, change in employment, is such that that a period of job loss would be a negative number).
In the majority of categories, we find noisy results, which is to be expected given the broad range of disease conditions falling within each category. However, two disease categories in particular are consistent with our thesis of demand pull-forward during periods of intense job loss, namely arthritis / orthopedic events (consistent with the idea that time-discretionary ‘events’ are more likely to be pulled forward) and neurology events. A closer look within the neuro category shows that the subset of diagnoses with apparent demand pull-forward include relatively discretionary diagnoses such as sleep disorders (including insomnia), narcolepsy, chronic pain, and neuropathy – again consistent with our thesis.
More granular results within arthritis / ortho are presented in Exhibit 5; here we find two diagnoses in particular that appear to drive results: spondylosis / intervertebral disc disorders / other back problems; and – to a lesser extent – the catch-all other bone disease / musculoskeletal deformities. Back problems, in particular, account for the majority of pull forward that we find in the broad arthritis category – for every 1,000 net jobs lost per month in 2001 – 2003, we find an additional 621 episodes of care related to back problems. We have every reason to believe that greater numbers of arthritis / ortho and neurology diagnoses ultimately would have led to increased rates of orthopedic device implantation, though to be clear the MEPS data do not contain procedure-level details, so we cannot prove such a link using the MEPS dataset.
Importantly, despite the clear relationship between arthritis / ortho and neuro demand with job loss, we did not find other pull-forward relationships that we had expected – specifically, we did not find a demonstrable pull-forward in rates of hospital admission.
The Company Reported View
Company reports offer a more up-to-date view than MEPS, but far fewer data points. We compared change in company-reported US sales with change in employment for those sub-sectors where we would expect a job-loss driven demand pull-forward. We made the comparison across three distinct job loss periods: 1Q01 – 4Q03, 1Q07 – 3Q08, and 4Q08 – 2Q10. Obviously, the latter two periods run sequentially, however the acceleration of the economic downturn associated with the end of 3Q08 Lehman bankruptcy creates two rather distinct patterns (rates) of job loss across the 1Q07 – 2Q10 period, thus the need to consider the pre- and post-Lehman periods independently.
Again, orthopedics shows the strongest fit with our thesis; in each time period (Exhibits 6 and 7) we find the expected relationship between job loss and demand. The case for hospitals is much weaker; we see the expected relationship in the more recent job loss period(s) (Exhibits 8, 9), but (consistent with the MEPS finding) do not see the relationship in the earlier (1Q01 – 4Q03) timeframe. We cannot show the expected relationship between job loss and US sales for commodity suppliers in any of the three job-loss periods (Exhibits 10, 11). The case for labs and dental also is quite weak, though this may simply be due to the very small number of observations. For labs, the relationship is shaped as expected but very weak in the most recent periods, but no relationship can be shown at all in the 1Q01 – 4Q03 period (Exhibits 12, 13). For dental supplies, the recent periods show no clear relationship, despite having shown the expected relationship in the earlier 1Q01 – 4Q03 period (Exhibits 14, 15).
The (Apparent) Consensus View
We compared changes in consensus revenue and EBIT expectations across a period beginning in June of 2009 (the last period before reported results would be impaired by having to compare against the most substantial (3Q08 thru 2Q09) period of job loss), and ending September of this year – the logic being that results throughout this period should have suffered from job-loss-inflated comparables. With the exception of PBMs (whose EBIT expectations we believe fell on expectations of gross-margin as opposed to utilization pressures), revenue and earnings estimates fell most substantially for biotech, device innovators (e.g. ortho and cardio), and non-innovative suppliers (e.g. BAX, BDX, COV, BCR) (Exhibits 16, 17). Our sense is that these revisions reflect an expectation of long-term / structural utilization pressures. We believe that the utilization pressures we’re seeing in current results are going to be short-lived, and so suspect that these revisions, if anything, are likely to be over-done.
We also tracked the change in relative price / earnings multiples (each sub-sector relative to healthcare) across the period. For those utilization-sensitive sub-sectors with falling sales and earnings expectations we also see falling P/E forwards relative to the rest of healthcare; to our minds this indicates that the buy-side’s interpretation of the utilization trend aligns with the sell-side, i.e. both estimate trends and stock prices indicate a belief that the present slowing of utilization is a new reality that will endure over the longer term.
The Bottom-Line on Utilization Trends, and Whether / Where They’re Exploitable
We obviously disagree with the market, and take the view that the utilization slow-down is a large but short-lived consequence of recent excessive rates of job loss. Accordingly we tend to favor sub-sectors that appear to have been re-priced on the assumption of a more sustained slowing of demand. Orthopedic names (e.g. ZMH, SNN, SYK, NUVA) show the tightest links between job loss and accelerated demand, and so appear to be the highest beta option for investing on the assumption of an improving volume trend. We see two primary risks with ortho: first, we see substantial odds of further price pressures in the export markets; and second, what our work says about the job-loss related slowing of demand – especially for spine – is distinct from what managements are saying. Managements are attributing much if not all of the spine slowdown to an increased frequency of claims denials, and while we’re skeptical of this interpretation, we cannot yet rule it out, and have to recognize that a permanently tightened claims environment would forestall or even eliminate meaningful recovery in ortho volumes. The more commodity-oriented suppliers (e.g. BAX, BDX, BCR, COV, HSP) have seen both similar revisions (Exhibits 16 and 17, again) and re-pricing (Exhibit 18, again) to the ortho names, and so offer a means of levering a portfolio toward a utilization recovery without facing any meaningful exposure to either tightening claims approvals or export market pricing pressures. Facilities offer an intuitive means of betting on a utilization recovery, but we stop short of such a recommendation, being unable to convincingly tie job loss to changes in utilization, or to show that the stocks have been meaningfully re-rated as a consequence of slowing utilization; and, because of other (largely bad debt related) concerns about the facilities investment case which we’ll explore in a later note. We similarly stop short of recommending biotech as a means of levering to improving volumes. Of the broad list of variables that create volatility in biotech stock prices, the underlying utilization trend is unlikely to dominate for any substantial period. And, we expect that the underlying concerns reflected in biotech’s recent revisions and share-price performance have less to do with overall (measurable) demand trends, and more to do with fears of a general (and far less measurable) tightening of biotech re-imbursement.
How Large is the Market for HDL-Raising Drugs?
MRK recently announced positive phase III (DEFINE) results for its lipid drug anacetrapib; in a 76-week study of 1,623 patients, anacetrapib reduced LDL-C (i.e. ‘bad’) cholesterol by 40%, and raised HDL-C (i.e. ‘good’) cholesterol by 138% (Exhibit 19). The results are promising in several respects; anacetrapib may be free of the blood pressure and morbidity effects that brought development of an earlier member of this class (CETP inhibitors, torcetrapib /PFE) to an end; and, anacetrapib appears to have a greater than expected positive effect on HDL-C.
Plainly anacetrapib faces a long and uncertain path to market, but the positive DEFINE results offer genuine hope that the CETP class may be commercially viable. We believe the potential market may be quite a bit larger than is generally appreciated.
It is exceedingly well established that lower LDL-C’s correlate with lower rates of cardio- and cerebro-vascular morbidity / mortality, and that the relationship holds whether subjects’ LDL-C’s are ‘naturally’ low, or lowered by drug therapy. It is similarly well established that ‘naturally’ high HDL-C’s are associated with lower rates of vascular events (Exhibits 20, 21); and, that the relationship holds even in the case of very low (< 70 mg/dl) LDL-C values that have achieved through drug therapy (Exhibit 22).
Nevertheless, in the absence of truly potent and tolerable HDL-C raising therapies, the question of whether raising HDL-C’s with drug therapy actually improves vascular event rates remains largely unanswered. For the sake of potential-market sizing, we’ll assume that anacetrapib and/or other HDL-C raising therapies establish such a beneficial relationship, and will also assume that tolerability is on par with currently available statins.
Exhibit 23 compares the LDL-C and HDL-C effects of currently available lipid regulating drugs with anacetrapib. Obviously the difference is anacetrapib’s dramatically larger effect on HDL-C; anacetrapib’s LDL-C effects are roughly on par with Zocor’s (simvastatin). Thus when we consider anacetrapib’s commercial potential, we generally view its LDL-C lowering capabilities as a commodity – though we recognize that anacetrapib’s LDL-C effects may have some non-trivial market appeal, simply by virtue of allowing patients to get to goal on lower (more tolerable) statin doses, without having to use ezetimibe (Zetia). Almost all (99%) patients can reach their prescribed LDL-C target on currently available therapies, and an increasing proportion of these (by end-2011, 77%) can do so on a generic (Exhibit 24).
Thus sizing the primary demand base for compounds like anacetrapib means identifying patients whose HDL-C’s are below presumed target levels – regardless of whether these patients have elevated LDL-C’s or are on statin therapy. Identifying the percentage of the population with various combinations of HDL-C, LDL-C and risk-factors is straightforward (Exhibit 25), but we can only guess at what may become the generally accepted HDL-C target(s).
Higher ‘naturally-occurring’ HDL-C levels appear to have an almost linear relationship with cardiovascular risks (Exhibits 20 & 21, again), i.e. each additional point of HDL-C above or below the population average (51.2 mg/dl) appears to correlate with similar changes in risk. Accordingly we have no clear means of sub-setting the population into patients who might or might not benefit from having a higher HDL-C — so we have little choice other than to make an educated guess, and have chosen to consider a range of HDL-C targets between 40 and 60 mg/dl. These values bracket the population’s mean HDL-C level; and, the higher target of 60 mg/dl is roughly the same ‘distance’ (about 20%) from the population HDL-C mean as the common LDL-C target (100 mg/dl) is from the population mean LDL-C of 123 mg/dl.
Despite the fact that LDL-C lowering will do little to generate sales for anacetrapib and/or other HDL-C raising therapies, we find the statin market a useful benchmark – the physicians and diagnostic screens are exactly the same, the value proposition for patients is generally the same; and, most candidates for HDL-C raising therapy also are candidates for LDL-C lowering. For reference, the global statin / ezetimibe market is roughly $35B (2009 sales); at its peak, the US market alone was almost $20B. In both the US and globally, lipid regulators (statins and ezetimibe) are the second largest therapeutic area. Current US prescription demand is just over 200 million annually; adjusting for compliance, this equates to just over 20 million persons in the US under treatment.
This (US) demand for statins is driven by the 43 percent of US adults having LDL-C’s above their ideal. 13 percent of US adults have both high LDL-C’s and HDL-C’s below 40 mg/dl. These US adults (equal to 30 percent of the statin-eligible population) presumably would be pre-diagnosed / treatment compliant candidates for HDL-C raising drug therapy. Add to these the 9 percent of US adults that have both normal LDL-C’s (i.e. they’re not going to be on a statin) and an HDL-C below 40 mg/dl, and a total of 22 percent of US adults would fall below this conservative HDL-C goal, and presumably would be eligible for HDL-C raising drug therapy (Exhibit 26). Thus under the conservative assumption of a very low HDL-C target, unit demand for HDL-C raising ultimately would be roughly half of statin unit demand.
Exhibit 27 follows the same logic for a more aggressive HDL-C target of 60 mg/dl (just 18 percent higher than the population mean). 35 percent of US adults have both a high LDL-C and a lower than 60 mg/dl HDL-C, i.e. just more than 80 percent of statin-eligible patients would be indicated for HDL-C raising treatment under a 60 mg/dl target. Add to these the 41 percent of US adults with normal LDL-C’s but below-target HDL-C’s, and we see that fully three-quarters of US adults would be eligible for HDL-C raising therapy under an HDL-C target of 60 mg/dl – nearly twice as many (1.8x) as are statin eligible.
In very crude terms, potential demand for HDL-C raising drug therapy is anywhere from very large to unprecedented, though we full recognize that the market would – like the anti-hypertensive and LDL-C lowering markets before it – tend to grow at moderate but sustained rates as the treatment base expands from higher risk (e.g. active coronary heart disease / very low HDL-C) patients to healthier patients with lower levels of immediate risk. Even at the conservative 40 mg/dl target (bear in mind the population average is 51 mg/dl), unit demand eventually would be roughly half of statin demand; such a therapeutic area today, at branded prices (approx. $3.50 / day), would be just over $10B in US sales alone, and would rank as the country’s fourth largest therapeutic area. Keeping in mind that our HDL-C targets are a guess, we’ll admit to having more faith in the higher (60 mg/dl) target, the simple logic being that if raising HDL-C reduces cardiovascular risks, and if the ‘wild-type’ linear relationship between HDL-C and risk-reduction remains intact, then it only makes sense to expect an (eventual) HDL-C target that is higher than the 51 mg/dl population mean.
Torcetrapib aside, we track 10 CETP-targeted / novel HDL-raising therapies that have entered the clinic (Exhibit 28).
Three of these have been discontinued, 2 from PFE and one from Avant, though we suspect the discontinuation of PFE’s two candidates (both in Phase I) has more to do with PFE backing away from the area in the wake of torcetrapib’s failure than with specific characteristics of either of these back-up compounds. In short, PFE’s back-ups could in theory re-enter development. With the exception of Roche / Japan Tobacco’s dalcetrapib, MRK’s anacetrapib appears to be the only CETP-inhibitor in Phase III, though anacetrapib’s apparent efficacy advantage over dalcetrapib appears sufficient to all but eliminate dalcetrapib as a meaningful threat (Exhibit 29). We recognize these efficacy comparisons are across distinct patient populations and are thus applies-to-oranges, though we suspect anacetrapib’s (particularly HDL-C) treatment effect has re-set the bar on what constitutes a viable development candidate, and so may force potential competitors – including at least some of those with current clinical candidates – back to their test tubes.
- Patients were able to report up to 4 diagnoses (which were translated into ICD-9 codes by professional coders) per event. Because codes are listed in the order they are mentioned by a patient, rather than prioritized, it is impossible to select one primary code as the reason for the visit. Thus, for events with multiple diagnoses, we attribute 1 full event to each diagnosis listed (rather than attribute a fraction of 1 total event to each of the multiple diagnoses). ↑
- We lag the employment explanatory variable to account for reporting delays and to allow time for employment signal processing by potential consumers. ↑
- Stryker(SYK), Zimmer(ZMH), Smith & Nephew(SNN), NuVasive(NUVA), Medtronic (MDT) ↑
- Universal Health Services(UHS), Community Health Systems(CYH), Lifepoint Hospitals(LPNT), Tenet Healthcare(THC), Health Management Associates(HMA), Select Medical (SEM), AmSurg Corp (AMSG) ↑
- N.B.: ‘commodity’ is relative here, and intended to distinguish companies having greater supply-chain focus from companies with a more research-based focus: Baxter(BAX), Covidien(COV), Becton-Dickinson(BDX), CR Bard(BCR) ↑
- Quest Diagnostics(DGX), LabCorp(LH) ↑
- Patterson Dental(PDCO) ↑
- Estimates rose for facilities, however the larger facilities companies’ stated intention of adding facilities through acquisitions was a major contributor to positive revisions, so revisions do not accurately reflect same-store expectations ↑
- Using LDL-C targets from the NCEP ATP III Guidelines (updated), we analyzed the percent of NHANES III participants that could reach their goal on currently available therapies. NHANES III data allow us to identify the proportion of the population having various combinations of risk factors (and thus different LDL-C targets) ↑