 To assess formulary decisions by Part D plans for selected newly approved drugs.

 Formulary placement and restrictions were identified for 33 drugs in 8 therapeutic classes (antihyperglycemics, anticoagulants, antiplatelets, disease-modifying agents for multiple sclerosis [MS] and rheumatoid arthritis [RA], chronic obstructive pulmonary disease [COPD] drugs, antiepileptics, and antipsychotics) in 863 Part D plans with continuous CMS contracts between 2009 and 2013. Multivariable models estimated the impact of drug characteristics and Part D plan characteristics on probability of drug adoption and, for adopters, evaluated factors associated with months to adoption and requirements for prior authorization (PA) or step therapy (ST).

 First Part D formulary placements varied from 2 to 14 months post FDA approval. On average, 56.7% of plans placed each drug within 6 months and 64.1% placed within 1 year of the National Drug Code assignment date. The most rapid adoption was for antipsychotics and antiepileptics. The slowest was for COPD drugs. More than 90% of disease-modifying agents for MS and RA were subject to PA. ST was uncommon except for antihyperglycemic agents. In adjusted analyses, enhanced benefit plans had a 4% higher probability of formulary placement (

<.01), and each additional star in the CMS star rating system increased the probability of adoption by 4% (

<.01). Overall, Medicare Advantage prescription drug plans had higher placement rates due to greater reliance on enhanced plan offerings and higher star ratings.

 We found significant heterogeneity in formulary placement and restrictions for 33 new drugs in the Part D marketplace between 2009 and 2013. Further research is necessary to determine whether this pattern applies to other drug classes.

Medicare spending per fee-for-service beneficiary in 2008 varied from $6000 in Rapid City, South Dakota, to more than $18,000 in Miami, Florida. While some policy makers favor directly reducing payments in high-cost areas, the Institute of Medicine favors policies focused on provider inefficiency. We investigated whether several such policies would have the virtue of also reducing variation in spending. If inefficiency is concentrated in high-cost areas, then this may be the case.

There are concerns that retail clinics provide an inferior quality of care, but our results do not support these concerns.

No difference was found in quality of antibiotic prescribing among retail clinics, physician offices, and emergency departments.

The Medicare Part D program provides drug benefits to more than 37 million Medicare beneficiaries enrolled in about 2700 plans operated by more than 150 sponsors.

 Plan coverage is guided by CMS formulary provisions that generally require plans to offer at least 2 products in each therapeutic class. Exceptions are made for innovator classes with just a single product and the so-called “protected classes” (immunosuppressants used for organ transplant rejection prophylaxis, antidepressants, antipsychotics, anticonvulsants, antiretrovirals, and antineoplastics) in which essentially all drugs must be covered. These exceptions notwithstanding, CMS-mandated therapeutic classes are broadly defined, giving plan sponsors considerable latitude in selecting products for formulary placement. Moreover, plans can influence utilization for covered products through cost-sharing tier assignment and utilization management tools, including step therapy (ST) and prior authorization (PA). As a result, there is a high degree of heterogeneity in benefit designs across the Part D market,

 which makes it extremely difficult to assess whether (and which) plans consistently offer coverage for high-value medications.

The objective of this paper was to improve policy makers’ understanding of how quickly Part D sponsors add newly FDA-approved drugs to their formularies. We had 3 specific aims. The first was to track rates of formulary placement by Part D plans for each new drug selected for analysis for a minimum of 12 months following FDA approval. Our second aim was to identify policies that plans adopted upon formulary placement, including ST and PA. Our final aim was to assess how formulary adoption decisions were influenced by characteristics of the drugs, number of competing products, and Part D plan characteristics.

Our approach exploits a unique characteristic of the Part D market in which stand-alone prescription drug plans (PDPs) are at risk only for drug spending, while managed care Medicare Advantage prescription drug plans (MAPDs) are also at financial risk for Part A and B services. We expected these differences in risk exposure to influence the timing of formulary adoption and the application of PA and ST and, ultimately, to lead to reduced beneficiary access to newly approved therapies in PDPs relative to MAPDs. Moreover, because MAPD plans are permitted to use savings generated in the provision of Part A and B services to subsidize Part D coverage, we expected to observe higher formulary placements among MAPDs. Findings of 2 studies suggest that MAPDs and PDPs respond to different incentives when constructing their drug formularies,

 but no published studies have examined formulary adoption of newly approved drugs in the Part D marketplace.

We selected all new drugs approved by the FDA between January 2009 and December 2013 in 8 therapeutic classes and then tracked Part D formulary coverage for these drugs through December 2014. We included 2 classes with relatively few competing products—anticoagulants and antiplatelets&mdash;as well as 2 classes with many competing products&mdash;antihyperglycemics and medications used in treating chronic obstructive pulmonary disease (COPD). We also selected 2 classes dominated by new, expensive biologic agents&mdash;disease-modifying agents used in treating multiple sclerosis (MS) and rheumatoid arthritis (RA)&mdash;and 2 CMS protected classes&mdash;antiepileptic drugs and antipsychotics. In all, we evaluated 33 drugs among the 438 products approved by the FDA during this time period.

 Although the drug selection criteria were qualitative rather than quantitative, we believe this study will help spur additional research on the diffusion of new drug products within the Medicare marketplace.

Data Sources, Sample Selection, and Study Variables

Primary data for the study were obtained from CMS monthly formulary files for all Part D plans from 2009 through 2014. The formulary files were supplemented with annual Part D plan characteristics from CMS Part D landscape files. Data on drug approvals and National Drug Code (NDC) assignments were obtained from the FDA.

To ensure consistent tracking of all time-related variables, we restricted the sample to 863 Part D plans with continuous CMS contracts over the study period (from 3728 plans in 2009 decreasing to 2660 in 2013). Continuity of contracts was determined based on unique identification (ID) numbers assigned to each Part D plan. Some plans had such short CMS contracts that we could not assess duration of formulary placement for most drugs of interest. Other contracts had gaps in coverage or changed sponsor-level names, making it impossible to accurately align formulary coverage and plan characteristics over time. The selected plans had a total Part D enrollment of 10 million in December 2009. By December 2013, enrollment in these plans reached 16.6 million, representing approximately 50% of total Medicare Part D enrollment that year.

The CMS monthly formulary files assign other unique ID numbers to each formulary. These numbers change every time there is a modification in formulary coverage, such as placing a newly approved drug. This means that, over time, all Part D plans have many formulary IDs. It was for this reason that we used the plan rather than the formulary as our unit of observation, recognizing that multiple plans may use the same formulary at any given time.

Study variables relating to timing of formulary placement included the FDA approval date, NDC assignment date, and calendar month in which each drug of interest first appeared on formulary among the 863 plans. In most cases, NDC assignment dates followed FDA approval dates by a few days, but in a few cases, the delay was 6 months or more. Because plans cannot place a drug on formulary without an NDC, we tracked Part D plan placement trends from the NDC assignment month. We measured the difference in months between the month the first Part D plan was observed to place the drug (“first adopter”) and the adoption month by each subsequent plan. We used first adopter date as the baseline, rather than FDA approval or NDC assignment date, because time prior to first adoption by any Part D plan was—by definition&mdash;the same for every adopting plan. The time horizon for measuring months to formulary adoption varied from a minimum of 12 months for drugs approved at the end of 2013 up to 5 years for drugs approved in early 2009.

Drug characteristics included drug type—new chemical entity (NCE), line extension (LE), or combination product (CP)&mdash;and timing of drug approval relative to other agents in the same pharmacologic class&mdash;first in class, second in class, or third or later in class. To determine place in class, we first identified the FDA Established Pharmacologic Class (EPC) for each drug of interest. We then searched Micromedex to identify all drugs within each EPC. Next, we identified FDA approval dates from the Drugs@FDA database.

 For plans that did place each drug on formulary, we captured ST and PA restrictions in the initial placement month from the monthly CMS formulary files.

Finally, we captured essential characteristics of plan structure and performance. On the structural side, we classified plans as either PDPs or MAPDs and determined whether each plan provided basic or enhanced benefits at time of NDC assignment. We also classified PDPs according to whether they offered premiums at or below regional benchmarks. We assessed each plan’s performance using CMS star ratings, with 5 stars representing the highest quality. We hypothesized that plans with enhanced benefits and higher star ratings would have earlier and higher formulary placement rates than basic benefit plans, benchmark plans, and those with lower star ratings. Unfortunately, we did not have information on manufacturer rebate offers and so could not model net cost as a variable in formulary placement decisions.

Our descriptive analyses consisted of drug-by-drug tabulations of FDA approval and NDC assignment dates, months to first Part D plan formulary placement, proportion of plans adopting each drug 6 months and 12 months following first placement, and ST and PA restrictions at time of initial formulary adoption.

We employed 2-part regression models to estimate the impact of drug and plan characteristics on formulary placement (part 1) and we identified factors expected to be associated with placement (months to formulary adoption and ST and PA restrictions) for plans that added the drugs to their formularies (part 2).

To assess factors associated with formulary placement (part 1), we created a dataset of 28,479 observations (863 plans × 33 drugs), with values of 1 for plans that placed a given drug on their formularies during the study period and 0 for plans that did not. We employed a similar model structure to estimate the part 2 models. Here, plan/drug observations were restricted to plans whose formularies adopted the drugs of interest. The same strategy was used to estimate the effects of drug and plan characteristics on ST and PA restrictions. All models were estimated using ordinary least squares regression so that the magnitude of the estimated effects could be readily compared across the various models.

We note 2 reasons why the effective sample sizes for these regression models are actually smaller than the nominal samples. First, multiple plans used the same formulary in any given month. The minimum number of unique formularies among the 863 plans was 91 in March 2011; the maximum was 133 in November 2012. Second is the potential for coordinated behavior in formulary adoptions within the same Part D plan. We corrected for the effect of clustering on standard errors using the Robust command in Stata (StataCorp; College Station, Texas).

 presents characteristics of the 863 Part D plans included in the study sample. The plans were divided almost evenly between PDPs (427) and MAPDs (436). Together they represented 241 contracts among 102 different sponsors. The PDPs were just about evenly split between basic plans (215) and enhanced plans (212), but among MAPDs, enhanced plans predominated (389 vs 47). Benchmark plans were offered exclusively by PDPs. Plan quality measured by star ratings was substantially higher among MAPDs (a mean ranking of 3.81 stars vs 2.91 stars among PDPs).

 provides descriptive information on all drugs of interest. Most of the products were NCEs (26), with CPs (5) and LEs (2) limited to antihyperglycemic agents and drugs used in treating COPD. Among NCEs, there was a wide mix of agents that were first in class (12), second in class (5), or third or later in class (16).

NDC codes were assigned within a median of 6 days of FDA approval across all drugs reviewed. NDC assignments for every anticoagulant, antiplatelet, disease-modifying agent for MS and RA, and antihyperglycemic agent occurred within 9 days of FDA approval. However, for most COPD drugs, antiepileptics, and antipsychotics, NDC approval dates were delayed by several months (10 months in the case of ezogabine [Potiga], an antiepileptic medication). From the NDC assignment date onward, there were further delays of between 2 months (golimumab [Simponi], lacosamide [Vimpat], and asenapine [Saphris]) and 9 months (rivaroxaban [Xarelto], ticagrelor [Brilinta], indacaterol [Arcapta Neohaler], and fluticasone/vilanterol [Breo Ellipta]) before first observed formulary placement. Both the mean and median times between NDC assignment and first formulary placement was 4.6 months. Within 6 months of first placement, 56.7% of the 33 drugs had been placed on the formularies of the 863 plans under review (47.5% if drugs in protected classes are excluded). Uptake ranged from 17.7% for linagliptin (Tradjenta) to 100% placement for the antiepileptics and 2 of the 4 antipsychotics. The adoption rate at 12 months post NDC assignment was 64.1% (57.2% if drugs in protected classes are excluded). For 3 drugs, our observation period (January 2006-December 2014) was too short to observe formulary coverage 12 months post first formulary placement.

The final 2 columns in Table 2 present summary statistics regarding new formulary placements. The application of ST and PA restrictions differed widely by therapeutic class. Across all classes, 10.8% of plans required ST upon formulary placement versus 29.5% for PA. ST was rarely applied to anticoagulants (1.1%), antiplatelets (0.5%), MS drugs (0.2%), RA drugs (0.1%), COPD drugs (2.4%), and antiepileptics (5.1%), but was required for between 19% and 35% of all antihyperglycemic drugs and antipsychotics. PA was required by 89% or more of plans for all MS and RA drugs. Overall, antiplatelet drugs were the least restricted.

 provides a breakdown of differences in formulary adoption rates between PDP and MAPD plans. Overall, MAPDs had higher adoption rates for 28 of the 33 drugs at 6 months (59.0% vs 52.3%) and 25 drugs at 12 months (65.5% vs 60.0%). For 3 drugs, MAPDs formulary placement rates were more than double those of PDPs at 12 months. Teriflunomide (Aubagio) was placed by 66.3% of MAPDs but only 31.6% of PDPs, liraglutide (Victoza) was placed by 44.0% of MAPDs and 20.0% of PDPs, and prednisone DR (Rayos) was placed by 31.6% of MAPDs versus 13.2% of PDPs. MAPDs had higher rates of adoption for expensive biologic agents used to treat MS and RA. Drugs with significantly higher placement rates among PDPs included prasugrel (Effient) (91.4% of PDPs at 1 year versus 78.4% of MAPDs), saxagliptin/metformin (Kombiglyze XR) (88.6% vs 76.9%), and linagliptin/metformin (Jentadueto) (84.8% vs 61.1%).

 summarizes results from the regression analysis. As hypothesized, enhanced benefit plans were significantly more likely to place new drugs on formulary (model 1) compared with basic benefit plans by 4 percentage points (

<.01). At the same time, enhanced benefit plans were 2% more likely to impose ST (

<.01) restrictions (models 3 and 4, respectively). Benchmark plans were less likely to add newly approved drugs (—4%; 

 <.01), and when they did, they delayed adoption (model 2) by nearly a month (0.89 months; 

 <.01) compared with nonbenchmark plans. Plans with higher star ratings were also significantly more likely to place new drug products per additional star (4%; 

<.01). After controlling for other plan characteristics, MAPDs were more likely to impose PA (4%; 

<.01) compared with PDPs. However, the fact that MAPDs were much more likely to offer enhanced benefits and earn higher star ratings (Table 1) meant that, overall, MAPD plans had higher and earlier formulary placement rates across the 33 drugs under investigation.

Drug characteristics played a bigger role in formulary placement decisions than did plan characteristics. Compared with NCEs, LEs and CPs were much less likely to be placed on formulary (—36%; 

<.01, respectively). However, plans placing these products did so more quickly than with NCEs (—1.04 months; 

<.01, respectively). The timing of drug approval relative to other drugs in class had consistently significant effects on all outcome measures. Being second and third in class increased the likelihood of placement by 5% and 7%, respectively (

<.1), but delayed placement by 4.53 months (second in class) and 1.77 months (third or later in class) (

<.01). Later entrants were also significantly more likely to be subject to ST but less likely to require PA.

Our analysis of 33 new FDA drug approvals in 8 therapeutic classes between 2009 and 2013 identified 3 sources of delay in the diffusion of new products within the Medicare marketplace. First were potential delays in assigning NDC codes to each new product. Second was the delay before the first Part D formulary places the drug. Third was the time it takes for formulary placement to diffuse within the Part D marketplace following first placement.

Normally, delays caused by failure to promptly assign NDCs are minimal, ranging from a few days to a month. Under certain circumstances, however, it can be months before NDCs are assigned. Across all of the drugs in our study sample, the mean delay between FDA approval date and NDC assignment date was 35.6 days, but this was heavily weighted by long delays for the 3 antiepileptic drugs, due in part to contention about if and where these medications were to be scheduled as controlled substances. Perampanel (Fycompa) was eventually placed into Schedule III and both exogabine (Potiga) and lacosamide (Vimpat) into Schedule V.

Delays between NDC assignment date and first Part D plan formulary placement averaged 4.6 months across the entire sample, with wide variation by therapeutic class: 2 to 4 months for antiepileptics and antipsychotics and up to 9 months for other products. The early placement of antiepileptics and antipsychotics was likely due to their protected class status. Variation in timing of first placement for drugs in nonprotected classes is more difficult to explain. One might surmise that products considered either highly efficacious or having some unique clinical advantage would be quickly adopted by plan formularies, but if that were the case, one would also expect to see a strong correlation between first placement and subsequent rapid diffusion. There was no such correlation in our data. Overall, 56.7% of all drugs had been placed on plan formularies within 6 months after first placement and 64.1% within 12 months following first placement. These adoption rates were heavily influenced by drugs in protected classes. Removing these from the averages dropped the mean adoption rate to 47.5% at 6 months and 57.2% at 12 months. As expected, the rate of adoption was fastest for LEs and CPs, as plans typically place these products on formulary without formal pharmacy and therapeutic (P&T) committee review or a waiting period.

The role that Part D plan characteristics play in formulary adoption decisions largely met our expectations. Nonbenchmark plans and those offering enhanced benefits had higher uptake rates for new drugs, as did plans with higher star ratings. After controlling for differences in plan characteristics, we found similar formulary adoption rates and time to placement among PDPs and MAPDs. However, the fact that MAPDs were predominantly enhanced benefit plans with high star ratings—both significant predictors of higher formulary placement rates&mdash;meant that MAPDs placed more new drugs on their formularies than did PDPs: 58.0% versus 52.3% at 6 months and 65.5% versus 60.0% at 12 months. Higher adoption rates for MAPDs are consistent with observations made in prior literature.

 However, these earlier studies did not correlate placement rates with application of utilization management tools. The fact that MAPDs were more likely than PDPs to impose ST and PA restrictions means that patient access to some newer medications could be more limited in MAPDs.

These results should be interpreted in light of several caveats. Most important is the fact that we evaluated a relatively small nonrandom sample of all FDA drug approvals between 2009 and 2013. Although our sample included a few LEs and CPs, it was weighted toward NCEs. Samples with different proportions of NCEs, LEs, and CPs would produce different estimates of delays in formulary placements by Part D plans.

Second, we restricted the analysis to all new approvals in just 8 therapeutic classes, albeit representative of a diverse set of classes used in treating common chronic conditions. Nonetheless, we make no claim that our results apply to therapeutic classes we did not investigate. Further research is necessary to determine whether the patterns observed in our analysis apply to other drug classes.

Third, we lacked data on final manufacturer prices (transaction prices minus rebates) faced by plans making formulary decisions for new drug products. Health plans have relatively little bargaining power when considering agents that are first in class. Subsequent approvals increase competition and generally lower acquisition prices.

Fourth, we restricted the study sample of Part D plans to those with continuous contracts from 2008 through 2013. This restriction was necessary in order to compute delays in plans’ formulary adoption behavior, but it also meant that our results are not universally generalizable to all Part D plans over the study period.

Fifth, limitations in the CMS formulary files precluded any formal evaluation of generosity of coverage of newly approved drugs. The files contained tier assignment numbers. However, the interpretability of tier numbers was hampered by the fact that cost-sharing tier levels varied both across plans and within plans over time (ie, a tier number of 3 could represent either a nonpreferred brand in a 4-tier plan with a single generic tier or a preferred brand in a 5-tier plan with 2 generic tiers).

Finally, we did not consider the clinical effectiveness of new drugs for Medicare beneficiaries in plan decision making regarding formulary placements. CMS rules and conventional practice by P&T committees place clinical effectiveness at the forefront of formulary considerations,

 but even when following standardized protocols for new drug evaluations, individual P&T committees may come to very different conclusions. Moreover, under CMS regulations, P&T committee decisions are recommendations that may be overruled by plan sponsors. The result is a very heterogeneous pattern of formulary adoptions of newly approved drug products across the Part D marketplace.

We found significant heterogeneity in formulary placement and restrictions for 33 newly FDA-approved drugs in the Part D marketplace between 2009 and 2013. Further research is necessary to determine whether this pattern applies to other drug classes.

 University of Maryland Baltimore (BCS, SET, MC, AJ, KS, DQ), Baltimore, MD; Novartis Pharmaceuticals Corporation (ENO, CZ, YP, SA), East Hanover, NJ.

 Dr Stuart and Dr Tom report receiving grant support from Novartis Pharmaceuticals Corporation. Dr Obi, Dr Zacker, Dr Park, and Dr Arcona report being employed by Novartis. Dr Obi, Dr Zacker, and Dr Arcona report stock ownership in Novartis. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (BCS, SET, ENO, CZ, YP, SA); acquisition of data (BCS, AJ, KS); analysis and interpretation of data (BCS, SET, MC, AJ, KS, DQ, ENO, CZ, YP); drafting of the manuscript (BCS, AJ, DQ, ENO, CZ, YP); critical revision of the manuscript for important intellectual content (BCS, SET, MC, AJ, DQ, ENO, CZ, YP, SA); statistical analysis (BCS, KS); obtaining funding (BCS, CZ, YP, SA); administrative, technical, or logistic support (MC, AJ, ENO, YP, SA); and supervision (BCS, CZ, YP).

Articles

WEB EXCLUSIVE

There is significant heterogeneity in formulary placement and restrictions on new drug approvals in the Part D marketplace.

Placement of Selected New FDA-Approved Drugs in Medicare Part D Formularies, 2009-2013

 Asthma is a costly and variable disease necessitating routine population health monitoring. Insurance claims represent all paid pharmacy, diagnostic, outpatient, inpatient, and emergency care; however, current claims-based identification tools may be overly specific. We sought to determine how various definitions of asthma may improve detection of patients at risk of asthma exacerbations.

 A statistical analysis of private insurance claims for patients in a pediatric primary care network in Massachusetts.

 We performed a retrospective statistical analysis for patients aged 2 to 18 years with 3 years of continuous enrollment. Multiple potential definitions were constructed and tested on 2 years of data against their ability to identify patients having an exacerbation in the third year. Definitions tested utilized patterns of medication fills and visits billed with a diagnosis of asthma, wheeze, or cough. We calculated the sensitivity and specificity of each definition and constructed a receiver operating characteristic curve.

 In a cohort of 28,363 patients, a definition identifying patients with 1 or more clinician visits with a diagnosis of asthma or wheeze over 2 years was most efficient in detecting patients with an exacerbation in the subsequent year (sensitivity, 0.78; specificity, 0.84). When tested on the same cohort, the Healthcare Effectiveness Data and Information Set (HEDIS) persistent asthma criteria were less sensitive but more specific (sensitivity, 0.20; specificity, 0.99).

 Population health registries and quality measurement may benefit from using a claims-based definition of pediatric patients at risk of asthma exacerbations that is not as restrictive as the HEDIS persistent asthma criteria.

This analysis demonstrates that definitions other than the Healthcare Effectiveness Data and Information Set (HEDIS) persistent asthma specifications can identify patients at risk of asthma exacerbations with greater sensitivity and little loss of specificity.

Organizations may wish to identify and provide interventions for populations with varying levels of risk, depending on their resources.

The alternative definitions studied here offer guidance on how to efficiently identify larger populations of at-risk pediatric patients.

Population health registries and quality measurement may benefit from using a claims-based definition of at-risk asthma that is not as restrictive as the HEDIS persistent asthma criteria.

Asthma is a common and costly pediatric condition that has proven difficult to control despite the availability of efficacious treatments.

 In-office prescription management and education are insufficient to adequately achieve control in many patients with asthma.

 Home visitation, case management, and some population health management strategies are effective but costly and therefore should be targeted to the most at-risk patients. However, determining asthma risk is difficult. Many organizations and quality improvement programs use the National Committee on Quality Assurance’s Healthcare Effectiveness Data and Information Set (HEDIS) criteria to identify patients with persistent asthma.

 The HEDIS definition of persistent asthma uses insurance claims to identify patients who have emergency department (ED) or inpatient stays for asthma, are frequently seen in the outpatient setting for asthma, or fill multiple asthma medications. However, this definition of persistent asthma identifies just the sickest patients with asthma. Many more who do not meet this definition may still be at risk of exacerbation and could benefit from monitoring and proactive management.

To better focus health efforts on the population of at-risk children with asthma, a more inclusive claims-based definition of asthma is needed: one that identifies the majority of patients who will have an asthma exacerbation in the near future but does not overidentify patients at low risk, for whom monitoring and interventions may be wasteful. We pursued this definition by creating multiple claims-based definitions of asthma and determining the sensitivity and specificity of each definition to identify patients who went on to have exacerbations. We hypothesized that the varying test characteristics of numerous definitions would allow for the construction of a receiver operating characteristic (ROC) curve. Such a chart could help population health managers choose a definition of at-risk asthma that identifies patients with a reasonable probability of having an exacerbation without expending resources on low-risk patients.

The Pediatric Physicians’ Organization at Children’s is an independent practice association of more than 90 pediatric primary care practices affiliated with Boston Children’s Hospital in Massachusetts. We analyzed paid insurance claims data from 4 commercial insurance companies that share data with our organization for quality improvement purposes. Using this database, we identified all patients aged 2 to 18 years with continuous enrollment during the measurement year and preceding 2 years.

We developed a number of claims-based definitions of asthma and compared their performance with that of the HEDIS definition of persistent asthma. Elements involved in the construction of the novel definitions included: clinician-assigned diagnosis of asthma (

International Classification of Diseases, Ninth Revision

] codes 493.0x, 493.1x, 493.8x, 493.9x), wheeze (

code 786.2); medication fill patterns (HEDIS medication set asthma medication ratio)

; and prior exacerbations of asthma (defined as an admission with a primary diagnosis of asthma, an ED visit with a primary diagnosis of asthma, or an outpatient visit for asthma with an oral steroid prescribed).

 Definitions evaluated outpatient, ED, and inpatient claims for clinician-assigned diagnoses, and many were revised iteratively throughout the analysis to capture a range of sensitivities and specificities. Each definition under consideration was applied to the claims database using a 2-year lookback (2012-2013) and a single-year lookback (2013), and each evaluated just the first 10 diagnoses of any claim. Some definitions evaluated just the primary diagnosis, whereas others searched for diagnoses in any position. Patients meeting a particular definition’s criteria were then evaluated for an asthma exacerbation in 2014. The sensitivity and specificity of each definition for exacerbation detection was then plotted to create an ROC curve, with the area under the curve (AUC) and its 95% CI calculated for each point.

All analyses were performed in SAS version 9.4 (SAS; Cary, North Carolina). The project was approved by the Boston Children’s Hospital Committee on Clinical Investigation. A waiver of informed consent was obtained because the research could not practicably be conducted without access to the protected health information and the risk to individual patients was no more than minimal.

A cohort of 28,363 children with continuous enrollment from 2012 to 2014 was identified, with a median age of 11 years as of the anchor date, December 31, 2013; 49.2% of the cohort were female. Asthma exacerbations were found in 390 children (1.4%) during the measurement year (2014). The many potential definitions of patients at risk of an asthma exacerbation are shown in the 

, along with the discriminating characteristics (sensitivity, specificity, and AUC) of each definition.

The formal HEDIS definition of persistent asthma, which requires meeting the HEDIS definition in each of the 2 years prior to the measurement year (2012 and 2013 in our sample), identified a cohort of 434 patients (1.5% of the total population), 17.7% of whom went on to have an exacerbation in 2014 (sensitivity, 0.20; specificity, 0.99). Patients meeting this definition accounted for 77 of the 390 patients who experienced an exacerbation in 2014. A more relaxed version of these criteria, applying the inclusion criteria just to claims in the prior year (2013 in our sample), identified 751 patients, 13.8% of whom had an exacerbation in 2014 (sensitivity, 0.27; specificity, 0.98). All other definitions were examined similarly, and test characteristics were plotted on an ROC curve (

The AUC of each definition was significantly different from 0.5, and the definition identifying patients who had been given a clinician diagnosis of asthma or wheeze on 1 or more claims in the prior 2 years (definition No. 18) had the highest AUC, indicating an efficient balance of sensitivity and specificity. Definition No. 18 identified a population of 4822 individuals (17.0% of the cohort), 6.3% of whom had an exacerbation in 2014 (sensitivity, 0.78; specificity, 0.84). Patients meeting this definition accounted for 305 of the 390 patients experiencing an exacerbation in 2014. Definition No. 17 (1 or more clinician diagnoses of asthma in the prior 2 years) performed nearly as well, with a sensitivity of 0.75 and a specificity of 0.86.

Creating disease-specific patient registries for population health initiatives increasingly requires organizations to “think outside the problem list” to find patients at risk. This work demonstrates such an approach for pediatric asthma using a claims-based definition of asthma that can efficiently identify patients at risk of an asthma exacerbation with reasonably high sensitivity and specificity. Our definition No. 18, which identifies patients with 1 or more paid claims for a clinician visit with asthma or wheeze as a diagnosis in the past 2 years, had the largest AUC—substantially larger than that of the formal HEDIS definition for persistent asthma.

The HEDIS definition can be useful in defining the denominator for certain quality measures, such as the asthma medication ratio, or as a research definition, but it identifies just a small and very high-risk population of children with asthma and excludes many children who will go on to suffer an exacerbation in the coming 12 months. Indeed, in our sample, the formal HEDIS definition excluded 80% of children who suffered an exacerbation in the measurement year. The HEDIS-identified cohort of patients in this study made up just 1.5% of the entire population, far below the established population prevalence estimates for pediatric asthma, such as the CDC’s estimate of pediatric asthma prevalence nationally (8.6%)

 and even further below the school-reported prevalence in Massachusetts (10.9%).

 In order to approach such population estimates, a broader definition is needed. Our definition No. 18 had the largest AUC but identified a cohort representing 17.0% of the total population, well above the population estimates. An individual organization balancing the desire to identify all patients at risk of an asthma exacerbation with the need for judicious allocation of resources may choose to use an alternative definition that is more or less sensitive, depending on its resource availability.

Increasingly, organizations are taking a proactive approach to managing patients with asthma by reaching out to their families to optimize preventive strategies, educate about the importance of adherence to an action plan, promote influenza vaccination, and reduce environmental triggers, among other evidence-based strategies.

 The cost of proactively managing at-risk children with asthma is high. Nonetheless, if applied to the right population of children with asthma, costly interventions can provide a positive return on investment.

 Building an accurate registry of patients with asthma for population health management purposes is difficult due to the inherent unpredictability of the disease, however. A claims-based identification rule may be optimal, because relying on provider memory is inaccurate,

 and assuming that any documented respiratory complaint represents asthma overestimates the desired population.

Although any loss of specificity from the HEDIS definition means that a larger proportion of the patients identified will not actually go on to have an exacerbation, there are clear benefits to a more inclusive identification definition that promotes outreach to a larger but still substantially at-risk population. There may be long-term benefits that are difficult to quantify for families who receive asthma education and mitigate environmental triggers in the home, such as by reducing exposure to environmental tobacco smoke. Given that genetic factors contribute to the development of asthma,

 such interventions can also benefit siblings. Additionally, missing fewer school days and workdays benefits families in innumerable ways.

This study does not help clinicians determine which children with wheeze will go on to become patients with persistent asthma, and it was not designed to assess the economic impact of an expanded registry of patients with asthma. Using a claims database also has the limitation of relying on assigned visit diagnoses to summarize nuanced clinical situations and lacks valuable demographic information. Although our study used 

 codes because of the time period of our data, the current 

International Classification of Diseases, Tenth Revision

) convention offers a more refined spectrum of asthma codes that incorporate severity, which could more accurately determine a patient’s risk. Although this analysis used a 2-year lookback period, it is effectively a cross-sectional study. A longitudinal approach may further refine the performance of various definitions, but the time period in this analysis adequately accounts for seasonal variation, and analyses of more recent data were limited by the conversion to 

In order to monitor and further characterize the patients in our network with asthma at risk of exacerbation, we have begun to use a definition of asthma that identifies patients with asthma or wheezing as any diagnosis on a paid insurance claim in the preceding 2 years (definition No. 18). This definition may serve as a more appropriate criterion for inclusion into the denominator of certain quality measures for children with asthma compared with the HEDIS definition of persistent asthma. Work is ongoing to determine which characteristics of patients with asthma portend a greater likelihood of an asthma exacerbation and what the relative contributions of these factors are to a patient’s risk of exacerbation.

Pediatric Physicians’ Organization at Children’s (JH, EKT, LV), Brookline, MA; Division of General Pediatrics, Boston Children’s Hospital (JH, LV), Boston, MA.

 This study was funded by internal funds of the Pediatric Physicians’ Organization at Children’s. No outside funding was used for this study.

 The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (JH, EKT, LV); acquisition of data (JH, EKT, LV); analysis and interpretation of data (JH, EKT, LV); drafting of the manuscript (JH, EKT, LV); critical revision of the manuscript for important intellectual content (JH, EKT, LV); and statistical analysis (JH, EKT, LV).

 Jonathan Hatoun, MD, MPH, MS, Pediatric Physicians’ Organization at Children’s, 77 Pond Ave, Ste 205C, Brookline, MA 02445. Email: jonathan.hatoun@childrens.harvard.edu.

1. Soni A. Statistical brief #434: the five most costly children’s conditions, 2011: estimates for U.S. civilian noninstitutionalized children, ages 0-17. Agency for Healthcare Research and Quality website. meps.ahrq.gov/data_files/publications/st434/stat434.shtml. Published April 2014. Accessed February 7, 2017.

2. National Asthma Education and Prevention Program. Expert Panel Report 3 (EPR-3): Guidelines for the Diagnosis and Management of Asthma-summary report 2007 [erratum in 

. 2007;120(suppl 5):S94-138. doi: 10.1016/j.jaci.2007.09.043.

CDC’s National Asthma Control Program: An Investment in America’s Health

. Chamblee, GA: CDC; 2013. cdc.gov/asthma/pdfs/investment_americas_health.pdf. Accessed February 7, 2017.

4. Asthma medication ratio: percentage of members 5 to 85 years of age who were identified as having persistent asthma and had a ratio of controller medications to total asthma medications of 0.50 or greater during the measurement year. National Quality Measures Clearinghouse website. qualitymeasures.ahrq.gov/summaries/summary/49708?search=Asthma%2C%20unspecified%20type%2C%20with%20%28acute%29%20exacerbation. Published October 2015. Accessed March 9, 2017.

5. Schatz M, Zeiger RS, Yang SJ, et al. Persistent asthma defined using HEDIS versus survey criteria. 

6. HEDIS 2017. NCQA website. ncqa.org/hedis-quality-measurement/hedis-measures/hedis-2017. Accessed March 9, 2017.

7. Vernacchio L, Trudell EK, Muto JM. Correlation of care process measures with childhood asthma exacerbations. 

. 2013;131(1):e136-e143. doi: 10.1542/peds.2012-1144.

8. Summary health statistics: National Health Interview Survey, 2014: table C-1a. CDC website. ftp.cdc.gov/pub/Health_Statistics/NCHS/NHIS/SHS/2014_SHS_Table_C-1.pdf. Published 2015. Accessed February 7, 2017.

9. Asthma in Massachusetts. CDC website. cdc.gov/asthma/stateprofiles/asthma_in_ma.pdf. Published 2008. Accessed July 11, 2017.

Pediatric Asthma in Massachusetts 2008 - 2009

. Boston, MA: Massachusetts Department of Public Health; 2012. mass.gov/eohhs/docs/dph/environmental/tracking/asthma-08-09.pdf. Accessed July 11, 2017.

11. Kelso JM. Do written asthma action plans improve outcomes? 

. 2016;29(1):2-5. doi: 10.1089/ped.2016.0634.

12. Cloutier MM, Hall CB, Wakefield DB, Bailit H. Use of asthma guidelines by primary care providers to reduce hospitalizations and emergency department visits in poor, minority, urban children. 

. 2005;146(5):591-597. doi: 10.1016/j.jpeds.2004.12.017.

13. Woods ER, Bhaumik U, Sommer SJ, et al. Community asthma initiative: evaluation of a quality improvement program for comprehensive asthma care. 

. 2012;129(3):465-472. doi: 10.1542/peds.2010-3472.

14. Hoeksema LJ, Bazzy-Asaad A, Lomotan EA, et al. Accuracy of a computerized clinical decision-support system for asthma assessment and management. 

. 2011;18(3):243-250. doi: 10.1136/amiajnl-2010-000063.

15. Weinberger M, Abu-Hasan M. Pseudo-asthma: when cough, wheezing, and dyspnea are not asthma. 

. 2007;120(4):855-864. doi: 10.1542/peds.2007-0078.

Analysis of insurance claims reveals that criteria other than the Healthcare Effectiveness Data and Information Set (HEDIS) persistent asthma definition can identify more at-risk patients with reasonable loss of specificity.

Identifying Children at Risk of Asthma Exacerbations: Beyond HEDIS

 This exploratory study used outpatient laboratory test results from electronic health records (EHRs) for patient risk assessment and evaluated whether risk markers based on laboratory results improve the performance of diagnosis- and pharmacy-based predictive models for healthcare outcomes.

 Observational study of a patient cohort over 2 years.

 We used administrative claims and EHR data over a 2-year period for a population of continuously insured patients in an integrated health system who had at least 1 ambulatory visit during the first year. We performed regression tree analyses to develop risk markers from frequently ordered outpatient laboratory tests. We added these risk markers to demographic and Charlson Comorbidity Index models and 3 models from the Johns Hopkins Adjusted Clinical Groups system to predict individual cost, inpatient admission, and high-cost patients. We evaluated the predictive and discriminatory performance of 5 lab-enhanced models.

 Our study population included 120,844 patients. Adding laboratory markers to base models improved R

 predictions of costs by 0.1% to 3.7%, identification of high-cost patients by 3.4% to 121%, and identification of patients with inpatient admissions by 1.0% to 188% for the demographic model. The addition of laboratory risk markers to comprehensive risk models, compared with simpler models, resulted in smaller improvements in predictive power.

 The addition of laboratory risk markers can significantly improve the identification of high-risk patients using models that include age, gender, and a limited number of morbidities; however, models that use comprehensive risk measures may be only marginally improved.

Laboratory tests that are frequently ordered by physicians in outpatient practices contain valuable data for individual risk assessment.

Ranges of blood chemistries and hematology results define a set of model markers that have clinical face validity and potential utility for care management.

Adding the laboratory-based markers to risk levels derived from claims, prescriptions, and enrollment data improves the prediction of individual cost, the prediction of inpatient admission, and the prospective identification of high-cost patients.

For practices, a simple model that includes demographics and laboratory information may provide a basic tool to evaluate patient panels.

Most predictive models in healthcare have relied upon diagnosis information from health insurance claims or other administrative data. Such claims-based predictive models have been used extensively by health plans and government agencies for provider profiling and payment, underwriting, and prioritizing patients for care management.

 Although claims remain an important source of risk data, the widespread implementation of electronic health records (EHRs) and other clinical information technology systems offers a new source of data on disease severity and health status, as most EHRs contain information not captured in claims, such as laboratory values, vital signs, and clinical assessments.

In the inpatient setting, laboratory tests have been used to assess the risk of mortality across a range of conditions, including acute myocardial infarction, congestive heart failure, diabetes, ischemic and hemorrhagic stroke, pneumonia, and septicemia.

 These predictive assessments of mortality risk have incorporated blood chemistries, hematology, and blood gases. Predictive models for mortality performed better after adding laboratory risk markers, but similar models predicting 30-day readmission did not improve as much.

Another case for laboratory data has been made for case-mix adjustment of inpatient admissions using diagnosis-related groups (DRGs).

 Clinical laboratory results combined with inpatient administrative data incrementally improved the ability of DRGs to explain the length of inpatient stays; however, Medicare Severity DRGs and other DRG versions do not incorporate laboratory data for inpatient classification.

Laboratory tests can be powerful predictors among certain patient populations. For example, patients with diabetes who maintained reduced glycated hemoglobin (A1C) levels (ie, had better glycemic control) had lower annual costs than patients with higher levels.

The goal of this study was to develop and evaluate an approach for transforming common outpatient laboratory tests into risk measures that could be useful when added to population-level predictive models. Our objective was to determine result ranges for several candidate blood tests that were associated with increased costs in the year after the tests were performed. We hypothesized that certain ranges of component results from blood tests in the base year would be associated with higher healthcare costs and increased inpatient utilization during the subsequent year. We also hypothesized that laboratory risk markers based on component ranges would improve predictive risk models for these outcomes, including models with demographic and Charlson Comorbidity Index (CCI) risk markers and 3 models from the Johns Hopkins Adjusted Clinical Groups (ACG) system.

We obtained data from HealthPartners, Inc (Bloomington, Minnesota), a health insurer and large integrated delivery system. Its database contains structured EHR data, including encounter diagnoses and laboratory test results; administrative data that included benefit eligibility files; and claims data with 

International Classification of Diseases, Ninth Revision, Clinical Modification

) diagnoses, Current Procedural Terminology (CPT) procedure codes from inpatient and outpatient settings, and filled prescriptions with National Drug Codes from outpatient pharmacies. HealthPartners provided these data for patients who were receiving care at facilities owned by the healthcare system.

Our study population included 120,844 patients who were continuously enrolled in 2012 and 2013 and had at least 1 visit to 1 of 5 HealthPartners outpatient clinics in the Minneapolis-St. Paul metropolitan area in 2012.

To harmonize the coding of test orders across HealthPartners’ entities, we mapped the internal HealthPartners codes to Logical Observation Names Identifiers and Codes (LOINC). LOINC is a common language for identifying health measurements, observations, and documents, and it is commonly used for laboratory orders and findings.

The assignment of LOINC was a 2-step process. We first used the Regenstrief LOINC Mapping Assistant to suggest potential LOINC, which were turned over to a pathologist for final review in the second step.

 All laboratory tests that we selected for this study were mapped to LOINC.

Selection of Laboratory Tests and Creation of Risk Markers

We identified 23 blood chemistries and hematology counts from 4 test panels (ie, the basic metabolic panel, lipid panel, liver function tests, complete blood counts) and extracted A1C, alanine aminotransferase (ALT), albumin, alkaline phosphatase (ALK), aspartate aminotransferase (AST), bicarbonate, blood urea nitrogen (BUN), calcium, chloride, creatinine, glucose, hematocrit, hemoglobin, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), platelet, potassium, sodium, total bilirubin, total cholesterol, total protein, triglycerides, and white blood cell results from the EHR data. Our literature review suggested that these tests are commonly ordered in office-based clinical practices, and our study population confirmed this theory, as 49% had at least 1 result for any of the 23 tests in 2012. We extracted CPT codes for test orders from claims submitted to HealthPartners and confirmed that essentially all results for the tests of interest were present in the EHR data. We compared results with reference ranges for healthy persons,

 manually excluded implausible results that were extremely far outside the reference ranges, and selected the most recent results to create patient-level risk markers. Patient-level annual costs were calculated from claims incurred in 2013.

We used a 3-step process to develop laboratory-based risk markers. First, we conducted a regression tree analysis for each of the 23 tests to determine result ranges that were prospectively associated with increased annual costs using the caret package in R software.

A strategy of individually analyzing laboratory covariates has been used to discriminate the risk of inpatient mortality.

 We minimized the impact of high-cost claimants by truncating individual costs at $250,000, which was equivalent to the 99.9th percentile in the study population. Second, several regression tree analyses generated multiple result ranges, and we condensed them into ordinal “low” and “high” levels to create binary markers. High levels of risk consisted of low test results, high values, or both. To prevent model overfitting, we required the high-risk groups to contain at least 1% of patients who had results for a test. Third, we created binary markers for low and high risk levels. Tested patients had either a low- or a high-risk marker assigned to their test results. High risk indicated a potential for high cost in the future period; low risk indicated that a patient’s condition was nonsevere or under control or that the test was performed for diagnostic reasons. Patients who did not have tests had all markers set to 0 so that we were able to evaluate the joint impact of laboratory-based risk markers in the entire patient population.

Our prospective outcome measures were individual total annual claims costs, presence of inpatient hospitalization, and high-risk patient status, defined as being among the top 5% of claimants in 2013. We developed 5 predictive models for these 3 outcomes.

The demographic model included age and gender only; the Charlson model included age, gender, and 17 morbidity categories from the CCI, which is widely used to predict high-cost patients.

 The ACG system contributed 3 predictive models; all included morbidity markers based on diagnoses, and 1 included morbidity markers that were derived from prescription data.

 We first combined the demographic variables with 32 Aggregated Diagnosis Groups (ADGs; ie, types of morbidity from the ACG system) into an ADG model. Other researchers have validated a similar ADG model for predicting mortality.

 Second, we implemented the ACG-Dx predictive model with age, gender, 24 selected ACGs, and 116 Expanded Diagnosis Clusters (EDCs) for the patient population.

 ACGs are mutually exclusive groups that are based on patterns of ADG morbidity types and have similar resource use; EDCs are clinical groupings of diseases based on 

International Classification of Diseases, Tenth Revision, Clinical Modification 

codes. A patient may have multiple ADGs and EDCs. Third, we implemented an ACG-DxRx model that contained all markers from the Dx version and included 65 Rx-MGs (ie, morbidity types based on prescription data). ACG-PM predictive models have been validated for predicting high-risk patients and inpatient hospitalization.

We used the health system’s data to generate individual risk scores. The models were run with and without the laboratory markers included. Our objective was to evaluate the contribution of new laboratory markers to the performance of predictive models with different levels of complexity as indicated by the number of morbidity markers.

We calibrated models for costs using ordinary least squares (OLS) and generalized linear regression and chose to report the coefficient of determination (R

) for OLS models. Models for inpatient hospitalization and high-risk outcomes were calibrated using logistic regression, and we computed sensitivity, specificity, area under the receiver operating characteristic curve (AUC), and integrated discrimination improvement (IDI) statistics to quantify the improvement in discriminatory performance due to laboratory markers. We included IDI because the original models that included ACG system variables showed high AUC values; therefore, meaningful improvements in discriminatory power might not be captured by measuring only AUC. The properties of the IDI statistic are well understood, and this statistic is increasingly used to evaluate markers that are introduced into predictive models.

The study population consisted of 120,844 patients who had at least 1 ambulatory care visit at a HealthPartners-owned clinic in 2012. The mean (SD) age was 37.6 (19.2) years, 20.6% were younger than 18 years, 3.3% were 65 years or older, and 57.1% were women (

). Almost all patients (99.9%) had at least 1 type of morbidity recorded, with an average (SD) of 5.9 (3.4) ADGs, and 20.0% had at least 1 comorbid condition that was included in the CCI. Mean (SD) patient total annual claims costs were $5732 ($20,208), and 5.1% were admitted to a hospital in 2013.

We extracted test orders from all outpatient medical service claims and measured laboratory data completeness. A1C results were 92% complete; calcium results, 96% complete; and all other test results, more than 98% complete.

We determined whether any result ranges for the candidate tests were associated with increased costs in the year after the tests were performed. Our analysis indicated separations between “low-cost” and “high-cost” risk groups for 12 of the 23 tests. These 12 tests included sodium, chloride, bicarbonate, glucose, and calcium from the basic metabolic panel; total protein and albumin from the liver function tests; hemoglobin, hematocrit, and platelets from the complete blood count; and total cholesterol and LDL-C from the lipid panel. The other 11 laboratory tests—A1C, ALT, ALK, AST, BUN, creatinine, HDL-C, potassium, total bilirubin, triglycerides, and white blood cell results&mdash;did not show an association with low- or high-cost risk and were excluded from the predictive modeling.

High-risk group sizes ranged from 1% of patients with hematocrit results to 21% of patients with albumin results. The average costs in these 2 groups were $32,695 and $24,234, respectively. Other high-risk groups showed similar increased average annual costs. The cost separation between high- and low-risk groups ranged from $4943 for LDL-C to $23,492 for hematocrit (

Predictive Model Performance Improvement With Added Laboratory Markers

Laboratory markers increased the prospective R

 of the demographic model for costs more than 2-fold from 2.2% to 5.9%; the IDI measures for inpatient and top-cost claimant identification were 121% and 188%, respectively (

 for cost increased from 10.3% to 11.4%, and the identification of inpatients and top-cost claimants, as measured by the IDIs, improved by 40% and 14%, respectively.

Overall, ACG-PM models exhibited higher prospective R

 values and showed less improvement with added laboratory markers compared with the demographic and Charlson models. The ACG-Dx and ACG-DxRx models predicted 22.2% and 24.7% of cost variation, respectively. Laboratory markers added small improvements to predicting costs across all 3 ACG system models; the R

The lab-enhanced ADG model had an AUC of 0.820 and an IDI of 4.8% for the identification of top-cost claimants. Similarly, lab-enhanced ACG-Dx and ACG-DxRx models had AUCs (IDIs) of 0.835 (1.5%) and 0.847 (1.0%) for high-risk identification, respectively. For hospitalization predictions, the AUCs across lab-enhanced ACG system models ranged from 0.789 (ADG) to 0.799 (ACG-DxRx); IDIs ranged from 8.4% for the ADG model to 3.4% for the ACG-DxRx model.

We developed high-cost risk markers using commonly ordered outpatient laboratory test results and evaluated how these markers improved individual predictions of healthcare costs, hospitalization, and high-risk status. This analysis extends previous research that used laboratory test results to predict clinical outcomes, such as mortality and hospital admission.

 We explored the potential value of these new commonly available clinical data sources for population-based predictive models as applied to care management.

We transformed test results that were extracted from an outpatient EHR into risk markers that could be replicated in a health system; organizations should be able to derive risk thresholds that are similar to those used in this research. Our thresholds were outside the reference ranges for apparently healthy persons, although less extreme than those used in previous inpatient mortality models.

Among the basic metabolic panel, 5 of 8 candidate tests were associated with high-cost risk. Abnormalities in electrolytes (sodium, potassium, chloride, bicarbonate, BUN) can occur in patients with congestive heart failure and kidney disease, and both conditions are linked with higher costs.

 Although creatinine is used clinically to determine chronic kidney disease stages, we found an association between creatinine and higher costs for less than 1% of tested patients, which was lower than our threshold. Hyperglycemia, as demonstrated by elevated glucose, was associated with increased cost risk, but A1C showed no association. Some tests that are commonly used to stage disease or guide treatment (eg, creatinine and A1C) were not predictive of prospective cost in our analysis. A previous analysis found that these tests were associated with 5-year costs among patients with diabetes

; our results may be accounted for by the 2-year duration of this study.

Among tests from the liver function test panel, 2 of 6 tests were associated with high-cost risk. Low levels of total protein and albumin are linked with liver disease and malnutrition.

 Three components from the complete blood count were associated with high risk in the subsequent year. Low hemoglobin and hematocrit values indicate anemia, and low values of platelets can be diagnostic of thrombocytopenia. Among the 4 tests included in the lipid panel, low total cholesterol and LDL-C levels were associated with high risk. Several conditions, including cancer, may contribute to low cholesterol levels.

 We did not identify high total cholesterol and LDL-C levels, 2 clinical risk factors for coronary artery disease,

An evaluation of the added value of risk markers derived from ambulatory laboratory tests in the prediction of healthcare costs and identification of high-risk patients.

June 2018

Placement of Selected New FDA-Approved Drugs in Medicare Part D Formularies, 2009-2013

Identifying Children at Risk of Asthma Exacerbations: Beyond HEDIS

Assessing Markers From Ambulatory Laboratory Tests for Predicting High-Risk Patients

Satisfaction With Care After Reducing Opioids for Chronic Pain

Cost Sharing for Antiepileptic Drugs: Medication Utilization and Health Plan Costs

A Longitudinal Examination of the Asthma Medication Ratio in Children

Prevalence and Predictors of Hypoglycemia in South Korea

Initial Results of a Lung Cancer Screening Demonstration Project: A Local Program Evaluation

Simply Delivered Meals: A Tale of Collaboration

Physician Practice Variation Under Orthopedic Bundled Payment