San Francisco Division DJENEBA SIDIBE, et al., Case No. 12-cv-04854-LB
Plaintiffs, ORDER GRANTING MOTION TO CERTIFY CLASS UNDER RULE v. 23(B)(3) AND DENYING MOTION FOR SANCTIONS Re: ECF Nos. 719, 735, and 747 Defendant. In this putative class action, the plaintiffs — four individuals who paid for health insurance and two small companies who paid for health insurance for their employees — sued Sutter Health for its allegedly anticompetitive practices, in violation of the federal Sherman Antitrust Act, the California Cartwright Act, and the California Unfair Competition Law.1 The plaintiffs claim that Sutter uses its considerable market power in seven Northern California markets (the tying markets, where it is the only or dominant hospital) to force five health plans in four other geographic markets (the tied markets) to accept Sutter’s hospitals in the tied markets at Sutter’s dictated, supra-competitive prices, which the health plans then passed through to consumers (such as the 1 Fourth Amend. Compl. (“4AC”) – ECF No. 204. Citations refer to material in the Electronic Case plaintiffs) in the form of higher premiums.2 In the tied markets, normally there is competition that would drive prices down and boost competition, but Sutter’s alleged tying practices allow it to leverage its “must have” hospitals in the tying markets to force inclusion of its hospitals in the tied markets at its prices and foreclose competition.3 The court previously certified an injunctive-relief class under Federal Rule of Civil Procedure 23(b)(2) but denied the plaintiffs’ motion to certify a Rule 23(b)(3) damages class because they did not establish that antitrust injury and damages were subject to common proof and predominated.4 The plaintiffs filed a renewed motion to certify a Rule 23(b)(3) class.5 The court grants the motion and certifies the class (except for the period from 2008 to 2010). The plaintiffs’ proposed class is as follows: All entities in California Rating area 1, 2, 3, 4, 5, 6, 8, 9 or 10 (the “Nine RAs”), and all individuals that either live or work in one of the Nine RAs, that paid premiums for a fully- insured health insurance policy from Blue Shield, Anthem Blue Cross, Aetna, Health Net or United Healthcare from September 28, 2008 to the present. This class definition includes Class Members that paid premiums for individual health insurance policies that they purchased from these health plans and Class Members that paid premiums, in whole or in part, for health insurance policies provided to them as a benefit from an employer or other group purchaser located in one of the Nine RAs.6 The court’s earlier class-certification order summarizes the market for hospital services, the sales by hospitals of services to health plans, the plans’ sale of health insurance to consumers (either individuals or employers), how hospitals compete to attract health-insurance enrollees as patients, and Sutter’s alleged anticompetitive practices.7 The main issue for certifying a Rule 23(b)(3) class is whether the plaintiffs have shown a reliable method for proving how overcharges
2 Id. at 3–5 (¶¶ 2–8), 9–12 (¶¶ 28–36), 13–15 (¶¶ 40–45), 29 (¶¶ 86–87), 34–35 (¶¶ 109–12). 3 Id. at 4–5 (¶¶ 5–8), 11–12 (¶¶ 35–36), 30 (¶ 94), 33 (¶¶ 103–05). 4 Order – ECF No. 714 at 42–52. 5 Mot. – ECF No. 735. 6 Order – ECF No. 714 at 5. were passed through to class members through higher premiums.8 The next sections summarize (1) the plaintiffs’ previous damages methodology and (2) the current damages methodology. 1. Previous Methodology Previously, the plaintiffs contended that health plans passed on 100 percent of Sutter overcharges to consumers through higher health-insurance premiums.9 The court held that the plaintiffs’ expert — Dr. Tasneem Chipty — did not support a 100-percent passthrough because she assumed the rate (as opposed to showing it) and ran regressions (based on the assumption) that measured only the correlation between costs and premiums without accounting for other factors affecting passthrough rates (such as competition from rival health plans, including Kaiser Permanente).10 The additional analyses in her reply declaration did not show a 100-percent passthrough either.11 In short, while “premiums generally increase when . . . costs increase,” the plaintiffs did not establish that health plans pass on 100 percent of cost increases through higher premiums or show any methodology for proving antitrust injury or damages on a class-wide basis.12 They thus did not show that common issues predominate.13 2. Current Calculation of Antitrust Injury and Damages to Class Members The parties do not dispute that the plaintiffs have demonstrated a reliable method for proving overcharges to health insurers and instead dispute whether Dr. Chipty’s passthrough methodology 8 Sutter previously challenged whether the plaintiffs had a reliable method for proving overcharges to all health insurers but does not dispute now that in her new analyses, Dr. Chipty estimated overcharges and applied a common methodology to the five class health plans. Opp’n – ECF No. 761–2 at 7–8. 9 Order – ECF No. 714 at 14–15, 20–27, 47–50. 10 Id. at 44–50. 11 Id. at 48–49. 12 Id. at 47–50; see 07/02/2020 Tr. – ECF No. 811 at 130 (p. 130:14–19) (The court: I understand that you’re quarreling with the percentage that Dr. Chipty has assigned. But you’re not quarreling with the conclusion that some significant part of the costs are passed through, as they necessarily must, under the Affordable Care Act. Right? Sutter: No, that part, Your Honor, I agree.”); Willig Decl., Ex. P5 to Cantor Decl. – ECF No. 736-4 at 44 (¶ 65) (“As a matter of economics, it is not controversial that there will be some amount of medical cost pass-through to premiums in the aggregate.”) — to show that the Sutter overcharges result in increased health-insurance premiums — is a sound methodology for proving antitrust injury or damages on a class-wide basis. Dr. Chipty conducted a regression analysis of the relationship between premium prices and medical costs and calculated that the overall weighted passthrough rate is 98.86 percent (as opposed to her earlier 100 percent): (a) 102.31 percent for Anthem; (b) 97.89 percent for Blue Shield; (c) 83.11 percent for Health Net; (d) 106.97 percent for Aetna; and (e) 102.10 percent for United.14 She capped the estimates at 100 percent to be conservative, which yielded an overall weighted average of 97.16 percent and (converting the passthrough rate to dollars) damages of $465.00 million to $489.04 million from September 2008 to December 2017.15 Her regression analysis controlled for 14 variables including medical costs, competition, a regulatory indicator that serves as a proxy for HMO products, time-invariant differences across health plans, and a time trend to allow for general changes in market conditions over time.16 Her damages model reflected the actuarial principle that health plans set premiums to cover their costs and earn a profit “within the bounds of regulations and subject to the competitive conditions of the market.”17 She described the four steps of the model that she applied to reach her 97.16 percent passthrough rate: Step 1: “Using a near-complete set of inpatient claims data from each of the five Class Health Plans, I estimate a set of multivariable regression models to determine the percentage by which each Class Health Plan was overcharged on Sutter inpatient hospital claims, by year, by Sutter Damage Hospital. Where the available data do not permit overcharge estimation, I do not calculate premium damages.
14 Chipty Suppl. Decl. – ECF No. 735-2 at 8–9 (¶ 11), 19 (¶ 28). 15 Id. at 8–9 (¶ 11), 33–34 (¶ 52) (“As I explained above, the coefficient on cost — in a model that studies the relationship between log(premium) and log(all medical cost) — represents the pass- through rate on a percentage-basis: it measures the percentage change (not the dollar change) in premiums that results from a percentage change in cost. This is not the same as passthrough rate on a dollars-basis; the pass-through rate on a dollars-basis is the portion of the cost increase that is passed through to consumers.”) (emphasis in original). 16 Id. at 11 (¶ 14). Step 2: Applying the overcharge percentages estimated in Step 1, I compute the amount by which each of the Class Health Plans overpaid Sutter, across the Sutter Damage Hospitals, by line of business, RA, and year. These overpayment dollars represent an increase in Class Health Plan costs. Step 3: I then calculate the dollar increase in premiums paid by Class Members, to each Class Health Plan, using a weighted average pass-through rate based on my econometric analyses of MLR Data: Overpayment by each Class Health Plan to the Sutter Damage Hospitals x Pass-Through Rate. Step 4: I compute aggregate premium damages as the sum of premium damages across all Class Members.18 Dr. Chipty identified six features that guided her regression analysis of the relationship between the premium prices and the medical costs. First, there is only one link in the supply chain and only one passthrough, which makes her analysis simpler than the analysis in cases like In re Qualcomm Antitrust Litig., 328 F.R.D. 280, 302–04 (N.D. Cal. 2018), where there are equipment manufacturers, wireless carriers, distributers, and retailers, and passthrough rates necessarily differ at different levels in the supply chain.19 Second, health insurance pools medical risk, meaning, a premium does not reflect the actual cost of an insured’s medical care and instead reflects the average expected cost of care. Thus, one should not study the relationships between premiums and medical costs incurred by purchasers (disaggregated to the purchaser-transaction level) and instead should study the relationship between average premiums and costs over a collection of purchasers.20 One can reasonably study premiums and medical costs at the health plan-business-line level, as Dr. Chipty did.21 Third, a model should contain information on all medical costs, such as facilities costs, physical and pharmacy costs, and capitation payments (which are fixed payments to medical 18 Id. at 15 (¶ 20). 19 Id. at 22–24 (¶¶ 35–36). 20 Id. at 24 (¶ 37). providers for health-plan members, regardless of the medical services they use), and a data set thus should have information on all medical costs.22 Dr. Chipty’s previous analyses used what she termed “production data” (statewide data produced by the class health plans to California’s Department of Managed Health Care (“DMHC”) on Uniform Rate Review Templates (“URRT”) for small groups), which represented the premiums and “inpatient and outpatient claims experience” for the class health plans.23 It included information only for facilities costs, not all medical costs.24 In her new analyses, Dr. Chipty used Medical Loss Ratio (“MLR”) data, which are statewide data on premium revenues and all incurred medical costs that the class health plans report — separately by line of business (individual, small-group, and large-group lines) — to the Centers of Medicare and Medicaid Studies (“CMS”) to ensure compliance with the Affordable Care Act’s (“ACA”) MLR requirements.25 The MLR data (a) “reflect the actual premium and claims experience for all Class Health Plans and Class Members across all business lines in California, from 2011 to 2018[,]” (b) “are constructed from transaction-level data by the health plans themselves to comply with federal regulations[,]” (c) “are filed with the federal government as a requirement under the ACA” and are used by CMS in the “ordinary-course . . . as an input into its assessment of whether health[-]plan premiums meet the MLR requirements[,]” (d) “contain the information necessary to study pass-through [] premiums[,] and all medical costs[,]” and (e) “contain sufficient information to study the importance of competition from Kaiser and non-Kaiser health plans on pass-through.”26 The MLR data are more robust than the production data, but she used the production data to corroborate her analysis of MLR data.27 To conduct her regression analyses, Dr. Chipty collected the MLR data associated with the State of California for all available years, 2011 to 2018. The key data fields are (a) total premiums, (b) total incurred
22 Id. at 25–27 (¶¶ 40–42). 23 Id. at 11–12 (¶ 16). 24 Id. at 12 (¶ 16). 25 Id. at 10 (¶ 13), 35–36 (¶ 55). 26 Id. at 10–11 (¶ 13). medical costs, and (c) member months.28 The data thus reflect the premium experience for (a) over 99.9 percent of class-member months from 2011 to 2018 and (b) about 70 percent of class member-months from September 2008 to December 2017.29 Fourth, prices (here, the premiums) can be a function of medical costs and product attributes, such as the extent to which patients and their health plans share costs of medical care (through co- insurance, co-payments, and deductibles) and the extent to which “gatekeepers” (e.g. the HMO model) control access to medical care.30 Her regression model controlled for these variables.31 She measured the cost split between the health plan and the member through the cost measure in the regression model. For example, the more a health plan pays toward the costs of medical care and the less the member pays out of pocket, the greater are the costs incurred by the plan.32 She controlled for the rules that patients follow to access HMO and PPO products by controlling for the regulator (because health plans submit MLR reports to different regulators (HMOs only to the DMHC and PPOs to the DMHC and the California Department of Insurance (“DOI”)).33 Fifth, she controlled for a regulatory environment that limits health plans’ flexibility in setting premiums.34 Health plans are highly regulated to ensure (a) solvency (through the requirement that they maintain financial reserves), which means that they cannot set premiums too low relative to medical costs, and (b) affordability (through the MLR requirement that the percentage of premium revenue spent on medical care and healthcare-quality improvement is not less than 80 percent for individual and small-group lines of business and 85 percent for large-group lines of business).35 Sixth, she controlled for competitive conditions, including the health plans’ competition with Kaiser. Kaiser, which has 45 percent of commercially insured health-plan members in California, 28 Id. at 36–37 (¶ 56). 29 Id. at 37–38 (¶ 57). 30 Id. at 28 (¶ 45). 31 Id. at 28–29 (¶ 45). 32 Id. at 39 (¶ 61). 33 Id. at 29 (¶ 45), 39 (¶ 61). 34 Id. at 30 (¶ 47). is a closed system that does not contract with non-Kaiser providers and thus does not pay Sutter overcharges.36 Nearly all of Kaiser’s plan members are HMO members regulated by the DMHC.37 The class plans (about 90 percent of commercially insured non-Kaiser plans in 2016) also compete with each other in multiple geographic areas in California, operate under the same regulatory framework, follow the same actuarial principles in setting premium rates, and launched steered products that failed because of Sutter’s restrictive contracting practices.38 Dr. Chipty accounted for the differential level of Kaiser’s presence in each business line (individual, small-group, and large- group lines), included a DMHC fixed effect, and concluded that the class plans are likely to pass through costs at a higher rate when they face more competition from Kaiser.39 The dependent variable in Dr. Chipty’s passthrough regression is the logarithm of per- member, per-month premium, which she calculated (using MLR data) by dividing total premium revenue by total member-months and taking the logarithm.40 She describes the explanatory variables in her baseline specifications as follows: Logarithm of Per Member, Per Month Cost: This variable is constructed by dividing total incurred medical costs by total member-months and taking the logarithm. The coefficient estimates associated with this variable reflect the pass-through rate on a percentage-basis. In my analysis, I allow the pass-through rate to vary by Class Health Plan and, in some versions, to vary by line of business. Five Class Health Plan Fixed Effects: These five variables take the value one for observations associated with that Class Health Plan, and zero otherwise. These variables are flexible controls that account for systematic, time-invariant differences in premiums across the five Class Health Plans.
36 Id. at 31–32 (¶ 49). 37 Id. at 39–40 (¶ 62). 38 Id. at 31 (¶ 48). 39 Id. at 11 (¶ 15), 32–33 (¶ 49), 39–40 (¶ 62). Two Line of Business Fixed Effects: These two variables — one for the individual line of business and one for the small group line of business — take the value of one for observations associated with that business line, and zero otherwise. The coefficient estimates associated with the individual and small group fixed effects reflect the incremental effect, on the log of per member, per month premium, of each of those two business lines relative to the large group business line. Thus, the model implicitly accounts for the large group line of business by virtue of including fixed effects for the other two business lines. The business line fixed effects control for systematic, time-invariant differences across the products types that are sold across business lines. Moreover, as I explained above, these variables also control for differential effects of Kaiser’s presence across business line, because Kaiser’s share is highly correlated with business line. A Fixed Effect for DMHC-Regulated Subsidiaries: This variable [] takes the value one for observations associated with products that are regulated by DMHC, and zero otherwise. As I explained above, DMHC regulates all HMO products. As such, the DMHC fixed effect is a proxy for an HMO product type. In addition, Exhibit 3 shows that DMHC regulates nearly all of Kaiser’s products. To the extent that Kaiser products are closer substitutes to non-Kaiser products regulated by the DMHC, this variable will also control for the differential effect of competition from Kaiser. Linear time trend: This variable takes the value one in 2011, two in 2012, and so on, up to eight in 2018. The linear time trend accounts for systematic trends in the log of per member, per month premium not reflected in the other variables in the model.41 Dr. Chipty ran three versions of her baseline regression model where the estimated passthrough rates were allowed to vary by (a) class health plan, (b) line of business, and (c) class health plan and line of business. As discussed above, her passthrough estimates were uniformly high and statistically significant, at the one-percent significance level, and vary from 106.97 percent for Aetna to 83.11 percent for Health Net, with confidence levels around each of the five passthrough estimates that include 100 percent, indicating that the estimate is not statistically different from a 100-percent passthrough.42 When Dr. Chipty capped the estimates at 100 percent, the overall weighted average was 97.16 percent, and (converting the rate to dollars) the damages were $465.00 million to $489.04 million for the period from September 2008 to December 2017.43 Sutter’s econometrician is Dr. Robert Willig, and he argued that Dr. Chipty’s estimates are inflated based on her Kaiser analysis. Kaiser and non-Kaiser plans have common cost shocks (from factors such as wage or medical-cost inflation), but Kaiser is unaffected by Sutter’s costs. He surmised that health plans would pass on common cost shocks but not Sutter overcharges because passing on Sutter overcharges would disadvantage the plans when competing against Kaiser. To address the concern, Dr. Chipty ran two regressions, using MLR data, to (1) add Kaiser medical costs (for a given line of business and year) as an additional explanatory variable and (2) add Kaiser costs and cost interactions.44 The results did not substantiate Dr. Willig’s concern and instead (for example) showed higher premiums as Kaiser’s share increased.45 Dr. Chipty identified quantitative evidence that corroborated her analysis of the MLR data, including her new analysis of the production data, her original econometric analyses, and Dr. Willig’s analyses of passthroughs.46 She also identified corroborating qualitative evidence from (1) health plans showing that they pass through increased medical costs through increased premiums, (2) Sutter documents about its business model and the effect of higher and lower costs on premiums, and (3) actuarial testimony that the class health plans set premiums to cover medical costs.47 Finally, she addressed several examples — both in the court’s earlier order and other 42 Id. at 44 (¶ 67). 43 Id. at 8–9 (¶ 11), 19 (¶ 28), 33–34 (¶ 52) (“As I explained above, the coefficient on cost — in a model that studies the relationship between log(premium) and log(all medical cost) — represents the pass-through rate on a percentage-basis: it measures the percentage change (not the dollar change) in premiums that results from a percentage change in cost. This is not the same as passthrough rate on a dollars-basis; the pass-through rate on a dollars-basis is the portion of the cost increase that is passed through to consumers.”) (emphasis in original), 47 (Ex. 6). 44 Id. at 47–48 (¶ 71). 45 Id. at 50 (¶ 73). 46 Id. at 51–68 (¶¶ 75–95). examples — of health plans giving rate passes (meaning, they did not pass on medical costs through premiums) to remain competitive.48 She said that (1) four examples of rate passes do not negate passthrough, (2) lower costs (without Sutter’s overpricing) would result in greater discounts and lower premiums, and (3) rate passes are uncommon in any event.49 Sutter submitted declarations from its experts Dr. Willig and Patrick Travis (from Deloitte Consulting) criticizing Dr. Chipty’s analysis on the grounds that the MLR data (1) are not representative of California consumers, (2) are aggregate data that do not capture variability (attributable to differences such as carriers, lines of business, HMO and PPO products, time periods, and rating areas), and (3) do not address the 2008 to 2010 time period, and Dr. Chipty does not account sufficiently for (4) Kaiser or (5) premium splits between employers and employees.50 In her supplemental declaration, Dr. Chipty observed that Dr. Willig did not challenge three points: (1) health plans pass through Sutter prices to plan members through health- plan premiums; (2) the production data from Health Net, Aetna, and United are unusable or unreliable, which is why she used MLR data; and (3) the passthrough is substantial, regardless of Kaiser’s competitive presence (discussed below).51 She then refuted the other criticisms. First, in response to the criticism that MLR data are not representative of California class members, CMS requires plans to file data in the states where their contracts are “‘sitused,’” which is significant — as data from the DMHC and DOI show — because “[a] comparison of the MLR database and the California Health Insurance Enrollment [d]atabase found that the commercial market enrollment totals from these two sources were within 5% of each other,” a gap possibly explained by small differences in reporting requirements.52 Also, the ACA allows interstate health plans, but there are not any.53 Dr. Chipty quantified the extent of cross-state enrollment using 48 Order – ECF No. 714 at 22; Chipty Suppl. Decl. – ECF No. 735-2 at 76–78 (¶¶ 105–06). 49 Chipty Suppl. Decl. – ECF No. 735-2 at 77–78 (¶ 106). 50 Travis Suppl. Decl. – ECF No. 761-4 at 6–18 (¶¶ 10–40); Willig Suppl. Decl. – ECF No. 761-6 at 57 (¶ 94), 67–68 (¶¶ 115–117). 51 Chipty Suppl. Reply Decl. – ECF No. 782-2 at 6–7 (¶ 8). 52 Id. at 36–37 (¶¶ 51–52). production data from Health Net, which had information on subscriber and employer location for each line of business, and found the following percentages of members living in California: 99.9 percent of individual-plan members, 98.7 percent of small-group members, and 98.5 percent of large-group members.54 Consistent with this showing that the level of out-of-state membership is small, there are published studies by healthcare researchers using California MLR data to study the California market.55 Second, contrary to Dr. Willig’s contention, Dr. Chipty did not assume a uniform passthrough rate and instead estimated average passthrough rates using appropriately aggregated MLR data.56 (She reiterated that MLR data are better than production data for the reasons in her first declaration, summarized above.57) Her averaging does not mask variation, and it is a judgment call whether variation is too great to undermine a class analysis. In her view, the class suffered a common impact.58 Passthrough rates are high and significant, and it is reasonable to average them.59 She did not ignore variation and made passthrough estimates using MLR data calculated by class health plans, class health plans by line of business, and line of business.60 As to variability across geographic areas, Dr. Chipty looked at all rating areas, not merely the class rating areas (where passthrough costs are higher), and thus it is conservative to average across all rating areas.61 She did not say that carriers pool risk on a state level, instead explained that they pool risk for individuals and small groups at the rating-area level and for large groups at the state level, and opined that one can analyze individual and small-group passthroughs on the state level (because variability is low across rating areas) and must analyze large-group passthroughs on the state level (because you cannot split a large group across a rating area without 54 Id. at 38 (¶ 54). 55 Id. at 38–39 (¶ 55) (citations omitted). 56 Id. at 7 (¶ 9). 57 Id. at 34–36 (¶¶ 48–50). 58 Id. at 7 (¶ 10), 13–15 (¶¶ 17–20). 59 Id. at 8 (¶ 11). 60 Id.at 15 (¶ 20). creating a mismatch of costs and premiums).62 Also, Dr. Willig’s presentation of passthroughs on a percentage basis gives a false impression of lower passthrough rates. When converted to dollars- based estimates, the passthrough rates are higher (with four out of five passthrough estimates for 2011 to 2014 that exceed 100 percent).63 Dr. Willig suggested splitting the MLR data into two subsamples (2011 to 2014 and 2015 to 2018), but there are not significant temporal differences, and in any event, averaging over all years is conservative.64 Dr. Chipty dismissed the argument that her passthrough analysis is likely overstated because of how health plans pool large (or catastrophic) claims statewide.65 Pooling risk is an analytic tool that does not change statewide medical costs or affect the reliability of the MLR data, and by studying the relationship statewide between per-member medical costs and premiums, she accounted for catastrophic claims under Dr. Willig’s approach.66 Dr. Chipty also dismissed the criticism that regulation by the DMHC or the DOI is not a proxy for HMO and PPO products: (1) 80% of DMHC’s regulation involves HMO products; (2) the DOI does not regulate HMO products; and (3) researchers acknowledge that DMHC regulates HMOs and the DOI regulates traditional health plans.67 Third, as to whether the MLR data can be extrapolated to 2008 to 2010, she contends that the data can be extrapolated to pre-ACA behavior because the health plans faced similar competitive conditions, “faced an MLR requirement, with California regulations governing pre-ACA and federal regulations governing post-ACA,” and used actuarial principles to set premiums.68 Fourth, she disputed Dr. Willig’s assertion that her analysis did not account for the disparity of Kaiser’s presence across the geographic rating areas and time periods.69 She controlled for (a) the 62 Id. at 9–10 (¶ 11), 15 (¶ 21). 63 Id at 8–10 (¶ 11), 16–20 (¶¶ 23–26). 64 Id. at 21–24 (¶¶ 30–32). 65 Id. at 39 (¶ 56). 66 Id. at 39–40 (¶ 56). 67 Id. at 40–42 (¶ 60). 68 Id. at 21 (¶ 29) regulator (DMHC) for Kaiser’s predominately HMO services, (b) the line of business (because Kaiser has a higher share in some lines of business), (c) time, and (d) geography. The MLR-data analyses account for the first three sources of variation, and the production-data analyses account for the fourth.70 The passthrough rate is “positive, highly statistically significant, and qualitatively similar, irrespective of how I account for Kaiser’s competitive presence.”71 “[T]here are generally no systemic differences in the passthrough rates in Kaiser-heavy and Kaiser-light areas.”72 Fifth, as to the proportional split of premiums between employers and employees, Dr. Chipty responded that accounting for the split is straightforward because a proportional split of the damage award between an employer and its employees, based on the actual cost-sharing arrangement, will exactly or reasonably compensate the class members.73 Class actions are governed by Federal Rule of Civil Procedure 23. A party seeking to certify a class must prove the prerequisites of Rule 23(a) and at least one subsection of Rule 23(b). The following are the prerequisites of Rule 23(a): (1) the class is so numerous that joinder of all members is impracticable; (2) there are questions of law or fact common to the class; (3) the claims or defenses of the representative parties are typical of the claims or defenses of the class; and (4) the representative parties will fairly and adequately protect the interests of the class. The court previously found that the plaintiffs proved all Rule 23(a) prerequisites.74 A court may certify a class under Rule 23(b)(3) if “the court finds that the questions of law or fact common to class members predominate over any questions affecting only individual members, and that a class action is superior to other available methods for fairly and efficiently adjudicating the controversy.” Fed. R. Civ. P. 23(b)(3). A court may certify a class under Rule
70 Id. at 25 (¶ 35). 71 Id. at 25–26 (¶ 36). 72 Id. at 26 (¶ 37). 73 Id. at 11 (¶ 13), 43–44 (¶ 63). 23(b)(2) for injunctive or declaratory relief if “the party opposing the class has acted or refused to act on grounds that apply generally to the class, so that final injunctive relief or corresponding declaratory relief is appropriate respecting the class as a whole[.]” Fed. R. Civ. P. 23(b)(2). The court previously certified an injunctive-relief class under Rule 23(b)(2) because the plaintiffs seek a single injunction barring Sutter from engaging in anticompetitive behavior.75 The issue in the renewed motion is whether the plaintiffs have met the prerequisites of Rule 23(b)(3). “[P]laintiffs . . . must actually prove — not simply plead — that their proposed class satisfies each requirement of Rule 23, including (if applicable) the predominance requirement of Rule 23(b)(3).” Halliburton Co. v. Erica P. John Fund, Inc., 573 U.S. 258, 275 (2014) (emphasis in original) (citing Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338, 350–51 (2011); Comcast Corp. v. Behrend, 569 U.S. 27, 32–33 (2013)). “[C]ertification is proper only if ‘the trial court is satisfied, after a rigorous analysis, that the prerequisites of Rule 23(a) have been satisfied.’” Comcast, 569 U.S. at 33 (quoting Wal-Mart, 564 U.S. at 350–51). “Such an analysis will frequently entail ‘overlap with the merits of the plaintiff’s underlying claim.’” Id. at 33–34 (quoting Wal-Mart, 564 U.S. at 351). “That is so because the ‘class determination generally involves considerations that are enmeshed in the factual and legal issues comprising the plaintiff’s cause of action.’” Id. at 34 (quoting Wal-Mart, 564 U.S. at 351). These analytic principles govern the court’s analysis under Rule 23(b)(3): “[i]f anything, Rule 23(b)(3)’s predominance criterion is even more demanding than Rule 23(a).” Id. (citing Amchem Prods, Inc. v. Windsor, 521 U.S. 591, 623–24 (1997)). Still, “Rule 23 grants courts no license to engage in free-ranging merits inquiries at the certification stage.” Amgen Inc. v. Conn. Ret. Plans and Tr. Funds, 568 U.S. 455, 466 (2013). “Merits questions may be considered to the extent — but only to the extent — that they are relevant to determining whether the Rule 23 prerequisites for class certification are satisfied.” Id. (citing Wal- Mart, 564 U.S. at 351 n.6). “Considering whether ‘questions of law or fact common to class members predominate’ begins . . . with the elements of the underlying cause of action.” Erica P. John Fund, Inc. v. Halliburton Co., 563 U.S. 804, 809 (2011). The claims here are antitrust claims under Sections 1 and 2 of the federal Sherman Antitrust Act, the California Cartwright Act, and the California Unfair Competition Law (“UCL”).76 “To establish a federal antitrust claim, plaintiffs typically must prove (1) a violation of antitrust laws, (2) an injury they suffered as a result of that violation, and (3) an estimated measure of damages.” In re Qualcomm, 328 F.R.D. at 296 (quotation omitted). The Cartwright Act — which was modeled on the Sherman Act — mirrors the analysis under federal law. Cf. id. (quoting Cty. of Tuolumne v. Sonora Cmty. Hosp., 236 F.3d 1148, 1160 (9th Cir. 2001)). The UCL claim “is premised at least in part upon the Sherman and Cartwright Act violations.” Cf. id. (citing Cel-Tech Commc’ns, Inc. v. L.A. Cellular Tel. Co., 20 Cal. 4th 163, 180 (1999)). “Neither party identifies any material difference between the federal and state claims warranting separate treatment.” Cf. id. “Thus, the [c]ourt may treat the state law claims together with the federal claims in this case.” Cf. id. The parties do not dispute Sutter’s allegedly anticompetitive conduct in violation of antitrust laws and dispute only antitrust injury and damages. “Antitrust ‘impact’ — also referred to as antitrust injury — is the ‘fact of damage’ that results from a violation of the antitrust laws.” Id. at 299 (quotation omitted). “‘It is the causal link between the antitrust violation and the damages sought by plaintiffs.’” Id. (quoting Brown v. Am. Honda (In re New Motor Vehicles Canadian Antitrust Litig.), 522 F.3d 6, 19 n.18 (1st Cir. 2008)). “Thus, Plaintiffs here must be able to establish, predominantly with generalized evidence, that all (or nearly all) members of the class suffered damage as a result of [defendant’s] alleged anti- competitive conduct.” Id. (quotation omitted). On a motion for class certification, “plaintiffs [must] be able to show that their damages stemmed from the defendant’s actions that created the legal liability” and “must show that damages are capable of measurement on a classwide basis, in the sense that the whole class suffered damages traceable to the same injurious course of conduct underlying the plaintiffs’ legal theory.” Nguyen v. Nissan N. Am., Inc., 932 F.3d 811, 817 (9th Cir. 2019) (quotation omitted). “[U]ncertainty regarding class members’ damages does not prevent certification of a class as long as a valid method has been proposed for calculating those damages.” Id. (quotation omitted). “Although uncertain damages calculations do not alone defeat certification, the Supreme Court has emphasized that ‘at the class-certification stage (as at trial), any model supporting a plaintiff’s damages case must be consistent with its liability case.’” Id. (emphasis in original, cleaned up) (quoting Comcast, 569 U.S. at 35). The plaintiffs’ liability theory is that Sutter charged the five class health plans — Blue Shield, Anthem, Aetna, Health Net, and UnitedHealthcare — supra-competitive rates. The alleged antitrust injury to class members is indirect: the class members’ harm is that the health plans passed through Sutter’s alleged overcharges to class members in the form of higher premiums for health insurance. The plaintiffs thus must demonstrate that (1) the five health plans paid Sutter inflated prices for inpatient hospital services, and (2) the overcharges were passed through to class members through inflated premiums. Cf., e.g., In re Qualcomm, 328 F.R.D. at 299. “[A]ntitrust plaintiffs have in recent years trended toward presenting an econometric formula or other statistical analysis to show class-wide impact and that such analysis has often been accepted at the certification stage.” In re Optical Disk Drive Antitrust Litig., No. 3:10-md-2143 RS, 2016 WL 467444, at *7 (N.D. Cal. Feb. 8, 2016) (In re Optical Disk Drive II) (quotation omitted). “[S]uch methods, where plausibly reliable, should be allowed as a means of common proof. To rule otherwise would allow antitrust violators a free pass in many industries.” Id. (quotation omitted). “Accordingly, it is clear that statistical and economic methodologies . . . may be employed to establish class-wide impact.” Id. (emphasis in original). Courts are cautious about “engaging in a battle of expert testimony” at the certification stage. Id. at *6 (citation omitted); accord, e.g., In re Lidoderm Antitrust Litig., No. 14-md-02521-WHO, 2017 WL 679367, *18 (N.D. Cal. Feb. 21, 2017). Still, “[c]ertification should not be automatic every time counsel dazzle the courtroom with graphs and tables.” In re Optical Disk Drive II, 2016 WL 467444, at *6 (quotation omitted). “If the presumption were otherwise, nearly all antitrust plaintiffs could survive certification without fully complying with Rule 23.” Id. (quotation omitted). “‘It is now purport to show predominance — the rule commands it.’” Id. (cleaned up) (quoting In re Rail Freight Fuel Surcharge Antitrust Litig. 725 F.3d 244, 255 (D.C. Cir. 2013)). “Put another way, the inquiry must be to determine if the proffered expert testimony has the requisite integrity to demonstrate class-wide impact.” Id.; accord Nguyen, 932 F.3d at 817 (plaintiff’s method for calculating damages must be “valid”). In her revised analyses, Dr. Chipty estimated overcharges and applied a common methodology to the class health plans.77 The parties dispute only the sufficiency of her analyses establishing the passthrough rates to the class members, who are indirect purchasers. When “the class is composed of indirect purchasers, proof of class-wide antitrust impact is made more complex because plaintiffs must offer a model of impact and damages that demonstrates the alleged overcharge was passed through to each successive link in the distribution chain, and ultimately to the plaintiffs.” In re Qualcomm, 328 F.R.D. at 301 (quotation omitted). Sutter challenges the plaintiffs’ passthrough model on the grounds that Dr. Chipty (1) inappropriately averages harm and relies on unverified MLR data and (2) does not identify a method for splitting premium increases between employers and employees.78 These objections do not preclude certification of the Rule 23(b)(3) class. Dr. Chipty’s analyses are tied to the plaintiffs’ liability theory within the meaning of Comcast and calculate the passthrough rates and resulting damages. The plaintiffs thus have shown that damages can be calculated on a class-wide basis. 1. Average Harm based on MLR Data Sutter contends that, while Dr. Chipty’s analysis recognizes that passthrough rates vary across carriers and lines of business, she ignores differences within lines of business and offers only “an approximation of the average harm . . . with a damages award that [] compensate[s] on average.”79 Sutter also criticizes her averaging over rating areas, her resulting failure to account for 77 Chipty Suppl. Decl. – ECF No. 735-2 at 8–16. 78 Opp’n – ECF No. 761-2 at 15–34. differences among rating areas, her failure to account for differences over time, and her extrapolation of the MLR data to 2008 to 2010, when MLR data were not available.80 Except for her extrapolation of the data to 2008 to 2010, Dr. Chipty accounts for these issues sufficiently. In its earlier order, the court summarized the relevant indirect-purchaser antitrust cases involving the use of averaging in calculating passthrough rates: Courts in indirect-purchaser antitrust cases have recognized the use of averaging in calculating passthrough rates where experts have shown the sound methodological steps through which they calculated their averages. See, e.g., [In re Qualcomm, 328 F.R.D.] at 315 (collecting cases). For example, In re Qualcomm — an antitrust case against a manufacturer of “modem chips” used in cellphones brought by a class of indirect purchasers who bought phones containing the chips — the plaintiffs’ expert calculated an average passthrough rate from the chip manufacturer through cellphone manufacturers and retailers to cellphone buyers through regression analyses based on sales data from (1) six major cellphone manufacturers, including the five largest manufacturers in the U.S. market (Apple, Samsung, Motorola, LG, and HTC), representing approximately 90 percent of total cellphone sales, (2) six of the largest U.S. retailers (including Best Buy, Amazon, Wal- Mart, and Target), representing approximately 84 percent of the retailer market, and (3) five wireless carriers, including the four major U.S. carriers (AT&T, Sprint, T-Mobile, and Verizon) and one regional carrier, representing approximately 97 percent of the wireless- carrier market. Id. at 302–03. The expert used ten control variables in his regressions (the same ten control variables that the defendant used in a submission to the FTC) and calculated separate passthrough rates for each of 18 sales channels, which he then weighted to calculate an overall sales-channel-weighted average passthrough rate of 87.4 percent. Id. at 303–04. In finding that the expert’s average weighted passthrough rate was sufficient to support class certification, the court found that: [The plaintiffs’ expert] does not simply assume a uniform pass-through rate for [original equipment manufacturers]. Instead, he examines transactional data for six different OEMs — including the five largest OEMs in the U.S. market (Apple, Samsung, Motorola, LG, and HTC) — who “accounted for approximately 90% of total cell phone sales” during the relevant period. [He] calculates individual pass- through rates for these six OEMs in order to model a composite pass-through rate. Id. at 308. By contrast, courts have rejected the use of averaging in calculating passthrough rates where experts have not shown the sound methodological steps through which they calculated their averages. For example, in In re Lithium Ion Batteries — an antitrust case against manufacturers of batteries used in consumer electronics — the plaintiffs’ expert assumed that the passthrough rate at each level in the distribution chain, from the battery manufacturers to device manufacturers to retailers to the end consumers, approached 100 percent without sufficiently supporting his analysis. In re Lithium Ion Batteries I, 2017 WL 1391491, at *12. Among other things, “[the expert] acknowledged that bundling, rebates, and discounts would affect the accuracy of cost data, but apparently has offered no methodology to account for it in his analysis.” Id.81 The court there “f[ound] the [expert’s] declarations insufficient to show that pass-through and damages can be established by expert analysis on a class-wide basis” and denied certification. Id. On a renewed motion a year later, the court found that the expert’s supplemental analysis still failed to account for a central issue affecting passthrough rates (focal-point pricing; meaning, the practice of retailers setting prices at certain “focal points,” such as prices ending with 9, and not adjusting such prices based on small differences in costs) and thus “left too much uncertainty as to whether pass-through can be estimated reliably at 100% as to retailers or distributors farther down the supply chain, and ultimately to the consumers who make up the proposed class,” and once again denied certification. In re Lithium Ion Batteries II, 2018 WL 1156797, at *4. The In re Optical Disk Drive antitrust litigation is instructive because the court there first denied class certification due in part to deficiencies in the plaintiffs’ passthrough assumptions and then, two years later, granted a renewed motion for class certification. The plaintiffs there alleged that optical-disc-drive (“ODD”) manufacturers engaged in bid- rigging to prop up the price of ODDs. In re Optical Disk Drive Antitrust Litig., 303 F.R.D. 311, 314 (N.D. Cal. 2014) (In re Optical Disk Drive I).[] The plaintiffs sought to certify a class of direct-purchaser plaintiffs (“DPPs”) that bought ODDs directly from the defendants, id. at 316, and a class of indirect-purchaser plaintiffs (“IPPs”) that bought products that contained ODDs manufactured by the defendants (but which the IPPs did not buy directly from the defendants), id. at 323. On the plaintiffs’ initial class-certification motion, the plaintiffs’ expert aggregated prices for all purchasers who bought ODDs of particular types in given years before running a regression analysis to calculate a passthrough rate. Id. at 324. The court found that this resulted in “class-wide impact . . . being assumed by the models, rather than demonstrated by the results.” Id. (emphasis in original). The expert “purport[ed] to test the validity of his models by looking to specific examples in the data,” but the court rejected this approach, holding that “[i]dentifying some instances where the empirical data appears to match the model does not transform the analysis from one that assumes class-wide impact into one that proves it.” Id. The court ultimately concluded that “the IPPs have not presented a persuasive explanation as to why it would be reasonable to assume a uniform pass through rate” and denied certification. Id. Two years later, the plaintiffs moved for certification again, this time for a narrower IPP class. In re Optical Disk Drive II, 2016 WL 467444, at *3. The plaintiffs’ expert offered a modified overcharge model that, among other things, “integrate[d] all ‘useable’ sales and costs data produced — from 86 percent of the market” and “provide[d] further detail on the multivariable regression analysis, which [the plaintiffs] contend shows that all factors other than [the antitrust] conspiracy are being adequately controlled for in the overcharge model.” Id. at *6. The court found that this new analysis, which measured passthrough rates for over 273 million ODD products, id. at *9, was sufficient to support certification.82
81 By contrast, the expert in In re Qualcomm “perform[ed] separate pass-through rate calculations for subsidized and unsubsidized phones and f[ound] statistically significant pass-through rates for each wireless carrier for subsidized and unsubsidized phones.” In re Qualcomm, 328 F.R.D. at 310. The court previously held that Dr. Chipty’s earlier analyses resembled the deficient passthrough models, not models that courts accept, because she assumed a 100-percent passthrough, supported that assumption with overly simplistic regressions, and ignored competition from rival health plans.83 Her new analyses cure these deficiencies. She did not assume a passthrough rate and instead calculated specific passthrough rates based on MLR data about total medical costs and medical premiums. She explained that she did not ignore variation and instead measured passthrough rates by class health plans, class health plans by line of business, and line of business. She examined all rating areas, explained why she averaged on a state level, considered the effect of Kaiser over the geographic rating areas and time periods, and corroborated her analysis of the MLR data with quantitative and qualitative evidence. She explained why it was reasonable to average the high passthrough rates, discussed why variation did not undermine her class analysis, and said that it was her judgment that the class suffered a common impact. In short, she has shown the sound methodological steps through which she calculated damages. In re Qualcomm, 328 F.R.D. at 315. This conclusion does not apply to 2008 to 2010 because MLR data do not exist for those years. (The MLR-data-reporting requirements began in 2011 after passage of the ACA.) Dr. Chipty applied her 97.16 weighted average to 2008 to 2010 on the grounds that the health plans faced similar competitive conditions, had reporting obligations under California law, and used actuarial principles to set premiums. The plaintiffs contend that this is an ordinary extrapolation from a data sample to a population.84 It is not. The intervening ACA required Dr. Chipty to evaluate data to determine whether passthrough rates remained constant. She did not. Moreover, her prior regressions relied on data for Anthem (the three lines of business) and Blue Shield (the small- group line), data that she rejects now in favor of MLR data and its superior capture of all medical costs (not merely facilities costs). The plaintiffs have not shown that damages can be calculated on a class-wide basis for 2008 to 2010.
83 Id. at 47–50. 2. Splitting Premium Increases Between Employers and Employees Sutter contends that the plaintiffs did not offer a viable method to allocate premium increases between employers and employees, generally because the inquiry is too individualized.85 This does not defeat certification of the class. First, the issue implicates allocation of damages, not calculation of damages, and thus does not affect the inquiry about whether the plaintiffs were injured at all. In re Lidoderm, 2017 WL 679367, at *4-5, *10–11 (insurers and insured together payed the full price for overpriced drugs, and insured were harmed even when they had co- payments). Second, any split of premium increases can be evaluated on a percentage basis based on the actual split of premium payments. 3. Other Arguments Sutter’s other arguments do not defeat certification of the Rule 23(b)(3) class. First, Sutter contends that the court must strike Dr. Chipty’s reliance on MLR data that the plaintiffs disclosed on November 18, 2019, after the deadline for expert disclosures.86 But she used new data because the court rejected her earlier methodology when it denied certification of a Rule 23(b)(3) class. There was no prejudice: Sutter responded with its own expert’s contradiction of Dr. Chipty’s model. The court denies the motion to strike the data and Sutter’s motion for sanctions based on the timing of the disclosure.87 Second, Sutter contends that Dr. Chipty’s model is novel, and she has not previously measured passthrough rates of medical costs to insurance premiums.88 The court’s previous conclusion — that Dr. Chipty has shown the sound methodological steps through which she calculated damages — disposes of these arguments. In any event, Sutter does not challenge her overall expertise. Dr. Chipty apparently has calculated passthrough rates in her consulting practice and is a principal 85 Opp’n – ECF No. 761-2 at 25–34. 86 Id. at 8–9; Opp’n to Sutter’s Sanctions Mot. – ECF No. 750-1 at 5. 87 Sutter’s Sanctions Mot. – ECF No. 747. 1 editor of the ABA treatise Proving Antitrust Damages, including its discussion of passthrough 2 calculations, and Dr. Willig apparently has not measured passthrough rates previously □□□□□□□□□ 3 5 The court grants the plaintiffs’ motion to certify a Rule 23(b)(3) damages class, except that it 6 denies the motion for the period from 2008 to 2010 (a period that precedes MLR data). The court 7 denies Sutter’s motion for sanctions. This disposes of ECF Nos. 719, 735, and 747. 9 Dated: July 30, 2020 Ld AC 10 LAUREL BEELER 11 United States Magistrate Judge
Q 16 & = 17
Z 18 19 20 21 22 23 24 25 26 27 28 *° Reply — ECF No. 782 at 4.