Methods of analyzing real world evidence (RWE) generally provide estimates that provide estimates for parameters of a population of patients, whereas the application of RWE to address decision making for the individual patient is less well established. The objective of this study was to review the literature to identify methods used for patient-level decision making using observational, real-world data sources. A systematic literature review was conducted in MEDLINE and PsychInfo. Eligible studies analyzed prospective or retrospective RWE, reported quantitative results, and described statistical methods applicable to patient-level decision making. Following dual eligibility review, details of the study design, methodology, strengths and weaknesses were extracted, verified by a second reviewer, and summarized quantitatively. A total of 115 articles met eligibility criteria (52 prospective, 22 retrospective, 37 cross sectional, and 4 multiple method studies). The most common statistical methods used in studies applicable to patient-level decision making included logistic regression (n=52,45.2%), followed by cox regression (n=24,20.9%), and linear regression (n=17,14.8%). Logistic regression was used mostly in prospective observational studies (25/52=48.1%) and in cross sectional studies (17/37=45.9%). Several studies included novel statistical approaches such as machine learning, recursive partitioning, and development of mathematical algorithms to predict patient outcomes. The strengths associated with these models included use of large datasets allowing both model development and validation in independent cohorts; however, few studies included external or cross-validation. Recommendations include validation in a different data source and limitations of variables collected. The studies identified in this literature search demonstrate a variety of rigorous statistical methods that can be used to develop patient-level decision-making models, tools and resources from large population-based datasets. Future observational research should consider these approaches for predictive modeling and should include both internal and external validation to ensure decision-making tools are accurate and reliable.

Methods of analyzing real world evidence (RWE) have traditionally focused on population-based outcome assessments, whereas the application of RWE to address healthcare decision making for the individual patient is less well established. The primary objective of this study was to systematically review published methods using real world data sources to inform patient-level and patient-provider decision making. A systematic literature review was conducted in MEDLINE and PsychInfo from 1/1/2000 to 9/18/2014. The search strategy included methodology, design and limited publications to cancer, diabetes, cardiovascular, Alzheimer’s disease, and rheumatologic conditions. A review of reference lists of identified articles enhanced the search strategy. Eligible studies were quantitative research that described statistical methodologies applicable to patient-provider decision making. Non-English and non-human studies were excluded. Articles were also excluded if they were qualitative research studies, reviews, policy/guidelines statements, or studies solely investigating a provider’s perspective. Following dual eligibility review, details of the study and research methodology were extracted and summarized. The search strategy identified 1088 publications. A preliminary review of 594 articles found 46 that were eligible. The methodologies used included prediction models based on logistic, multiple regression and Cox-regression models, multivariate risk analyses, discrete choice experiments, net reclassification, and classification trees. The review is currently ongoing and additional eligible articles will be identified. Details of the application of these methods to patient-provider decision making will be presented at the meeting. There is a need to incorporate methods in research studies to support evidence-based patient-level decision making. This systematic literature review has sought to identify methods and exemplars that will enable researchers to produce work to inform patient-provider decision making. This review will make recommendations regarding appropriate methods for scientists and investigators conducting patient-centered research. Future research should consider these approaches to incorporate methods that will support evidence-based patient decision making.

To compare units per day per claim (units) and costs per day per claim (costs) of comparable insulin products by Eli Lilly and Company (LLY) and Novo Nordisk (NN), adjusting for baseline patient differences, in state Medicaid claims data. Claims for comparable LLY or NN insulin for patients with continuous coverage for ≥6 months before their first observed insulin claim (baseline) were identified from Missouri (MO: 1/1/2011-3/31/2012) and New Jersey (NJ: 1/1/2011-3/31/2013) de-identified Medicaid claims data. Units were calculated by multiplying total quantity per claim (in mL) by strength (1 mL=100 units) and dividing by total days supplied. Costs were calculated (for patients aged <65 years only, because drug costs for those aged ≥65 years are often covered by Medicare rather than Medicaid) by dividing the cost of a claim to insurers by total days supplied. Regression-adjusted units and costs were estimated using generalized estimating equation models, accounting for baseline demographics, select comorbidities, and antidiabetic medication use. Claims for 23,325 MO and 9,749 NJ Medicaid patients were analyzed. Compared with NN insulin users, LLY insulin users were significantly younger, had lower rates of comorbidities, and higher rate of baseline insulin use. The regression-adjusted units for all comparable LLY and NN insulins were similar, with the exception of significantly lower units for insulin lispro (MO only: 67.6 vs. 73.2, P=0.0009) and LLY human insulin regular vials (MO: 65.4 vs. 78.3, P<0.0001; NJ: 45.3 vs. 50.3, P=0.0365). The regression-adjusted overall cost was significantly lower for comparable LLY vs. NN insulin (MO: $5.7 vs. $6.1, P=0.0046; NJ: $4.6 vs. $5.5, P<0.0001). In both MO and NJ Medicaid, the units of comparable LLY and NN insulins in years evaluated were similar for patients with similar characteristics; however, the overall cost was significantly lower for comparable LLY vs. NN insulin.

This US study aimed to determine average daily dosing of liraglutide from a national payer perspective. This study utilized 211,184 liraglutide pharmacy claims from the national Truven MarketScan® Database (10/1/2011-9/30/2012). Patients had type 2 diabetes (DM) diagnoses x 2, no type 1 or gestational DM, and were ≥18 years old (N=56,971). Liraglutide is typically dosed in 1.2 and 1.8 mg/day injections. DACON was based on total daily quantity prescribed (ml x 6 mg/ml ÷ days’ supply). A number of claim quantities and days’ supply values ranged from negative/zero to very high and implausible values, which translated to extremely high daily dose results (skewed distribution). The anomalies were addressed by conducting several analyses. The primary analysis included quantities from 6-27 ml/day and calculated mg/day DACON values as <0.6, 0.6-1.5=1.2, >1.5-2.1=1.8, and >1.8. Sensitivity analyses were conducted for: (A) all values, (B) quantity values “corrected” by corresponding price, (C) quantities corresponding exactly to 1.2 and 1.8 mg/day, (D) quantities with calculated DACONs between 0.6 and 1.8. Additional analyses were performed on excluded patients (N=30,098). On average, patients were 55 years old, 53% female, 60% from PPO plans. Comorbidities included hypertension (48%), cardiovascular disease (11%), obesity (10%), and neuropathy (9%). The DACON for primary analysis was 1.64 (34.4% at 1.2 mg/day; 64.2% at 1.8 mg/day). A sensitivity analysis of all positive claims (A) produced a DACON of 1.97 with 3% of claims >1.8 mg/day; additional DACON analyses were 1.63(B), 1.64(C), and 1.59(D). DACON primary analysis and all-value (A) analyses on liraglutide patients not meeting inclusion criteria were 1.65 and 1.95. Careful inspection of claims data distributions should guide methods used to arrive at reasonable, compelling findings for analyses even as simple as DACON. Liraglutide’s DACON in use with type 2 DM ranged from 1.59 to 1.64.

Topical testosterone agents (TTAs) are commonly used to raise low serum levels of testosterone in men. After initiating at a recommended starting dose (RSD), patients may undergo dose titration to achieve an appropriate maintenance dose. The objective of this study was to compare daily maintenance doses and costs of treatment with TTAs from US payer perspective in adult men diagnosed with Hypogonadism (HG). Adult men with a HG-associated diagnoses initiated at the RSD with Axiron® (Lilly USA, LLC; RSD 60mg per day; N=209), AndroGel® 1% (AbbVie Inc.; 50mg per day; N=614), AndroGel® 1.62% (AbbVie Inc.; 40.5mg per day; N=235), or Testim® (Auxilium Pharmaceuticals, Inc.; 50mg per day; N=558) between January 1,2011 and March 31,2012 were identified in a database of commercially insured beneficiaries. Patients were required to have continuous eligibility and no claims of the index therapy in 12 months prior to and at least 1 month of continuous eligibility following initiation. Baseline demographic characteristics, Charlson Comorbidity Index (CCI), comorbidities, and testosterone use were compared using chi-squared test for categorical variables and Wilcoxon rank-sum test for continuous variables. Mean dose was estimated per-person per-day (PPPD). Risk-adjusted dose and payer costs PPPD were estimated with a generalized linear model. Maintenance dose was attained at month 4. Mean dose PPPD in month 4 was 68.45mg, 56.68 mg, 51.18mg, and 59.24mg and risk-adjusted dose PPPD was 113.3%, 113.5%, 126.3%, and 118.7% of RSD for Axiron, AndroGel 1% (non-significant vs. Axiron), AndroGel 1.62% (P<0.001 vs. Axiron), and Testim (P=0.047 vs. Axiron), respectively. Risk-adjusted third-party payer costs PPPD were $7.49, $9.49, $10.39, and $9.59 (all P<0.001 vs. Axiron), respectively. Maintenance dose as a proportion of RSD was the least among Axiron and AndroGel 1% patients, while third-party payer costs for the maintenance dose were lowest among Axiron patients.

Patterns of care following topical testosterone agent (TTA) initiation are poorly understood. This study aimed to characterize care following TTA initiation and compare results between patients with and without a serum testosterone (T) assay within 30 days before and including TTA initiation. Adult men ( N = 4,146) initiating TTAs from January 1, 2011, to March 31, 2012, were identified from a commercially insured database. Patients were included if they initiated at recommended starting dose (RSD) and had ≥12 and ≥6 months of continuous eligibility preinitiation (baseline) and postinitiation (study period), respectively. Patients were stratified by preinitiation T assay. Maintenance dose attainment month was determined using unadjusted generalized estimating equations regression to compare dose relative to RSD month by month. Outcomes included maintenance dose attainment month, time to stopping of index TTA refills or a claim for nonindex testosterone replacement therapy (TRT), and proportion of patients with study period T assay or diagnosis of hypogonadism (HG) or another low testosterone condition, and were compared using chi-square and Wilcoxon rank-sum tests for categorical and continuous variables, respectively. Maintenance dose was attained in Month 4 postinitiation, at 115.2% of RSD. Approximately 46% of patients had a preinitiation T assay; these men were more likely to receive a diagnosis of HG or another low testosterone condition, to have a follow-up T assay, to continue treatment by filling a nonindex TRT, and less likely to stop refilling treatment with their index TTA. Differences in care following TTA initiation suggest that preinitiation T assays (i.e., guideline-based care) may be helpful in ensuring treatment benefits.

To evaluate if the Iowa Medicaid duloxetine depression Prior Authorization (PA) policy, implemented may 24, 2010, inadvertently increased atypical antipsychotic use in depressed patients. We compare initiations of duloxetine and other relevant medications for depression in Iowa before and after PA implementation and in Missouri, which had no duloxetine PA. Using de-identified Iowa and Missouri Medicaid claims data (1999-2010), two cohorts were selected from each state: 2010 policy change cohort (index date: 5/24/2010) and 2009 control cohort (index date: 5/24/2009). Patients had to have ≥1 inpatient or 2 other medical claims with a depression diagnosis pre-index; ≥1 antidepressant or antipsychotic claim during the six months pre-index (“baseline period”); and age<65 for six months post-index (“study period”). Baseline characteristics and study period prescription drug initiations (requiring six-month washout) by PA status (Iowa PA policy begun 5/24/2010) were compared between the two cohorts in each state. Logistic models were used to calculate risk-adjusted study period drug initiation rates, controlling for baseline characteristics. Iowa patients had significantly (p

To assess clinical and resource utilization outcomes between elderly type 2 diabetes patients initiating basal insulin analog administered via pens vs. vials. An online survey of 352 U.S. primary care physicians was used to collect retrospective patient chart data on 500 elderly type 2 diabetes patients who initiated on basal insulin analog in 2009. Information on glycemic control, severe hypoglycemic events, and diabetes-related health care resource utilization was collected prior to initiation and over a 12-month study period following insulin initiation. Information on adherence, both objective (measured via proportion of days covered [PDC]) and physician-revised (based on a revision of the objective PDC) was also collected for the study period. Multivariate regression analyses were used to control for confounding factors. Over the study period, initiation on pens, compared to vials, was associated with HbA1c levels 0.14 points lower (P=0.027). No differences were seen in the likelihood of reaching an HbA1c thresholds of 7% (hazard ratio [HR]=1.05, P=0.738), though the likelihood was 27% and 55% higher for pen initiators for HbA1c thresholds of 7.5% (HR=1.27, P=0.041) and 8% (HR=1.55, P<0.001), respectively. Rates of severe hypoglycemic events, inpatient, outpatient, and endocrinologist visits were similar between cohorts, but rates of emergency room visits were 68% lower for the pen cohort (IRR=0.32, P=0.017). No significant differences in objective adherence rates (PDC) were observed (P=0.080), while physician-revised adherence levels indicated a 2.2 percentage point higher adherence level for the pen cohort (P=0.034). Results from this study suggest that initiation on insulin pens, as compared to vials, may be associated with better glycemic control, fewer emergency room visits, and higher adherence levels (based on physician assessments) without an impact to the incidence of severe hypoglycemic events. Objective adherence and other resource utilization categories were not significantly different between the two cohorts.

To assess physician beliefs on the relative benefits of pens vs. vials, prescribing drivers, and characteristics of patients newly initiated on basal insulin analog via pens vs. vials among elderly type 2 diabetes patients. An online survey of 352 U.S. primary care physicians was used to collect retrospective patient chart data on 500 elderly type 2 diabetes patients who initiated on basal insulin analog in 2009. For each physician, patient chart selection was randomized among eligible patients. Data on physician characteristics, physicians' opinion on the impact of different routes of administration (ROA), main drivers for selecting a particular ROA for each patient, and patient characteristics were collected. The majority of physician respondents were part of a group practice (77.8%) and treated a mean of 235.5 (SD=185.0) type 2 patients ≥ 65 years old in 2009. Patient characteristics were similar in terms of age and diabetes duration. However, significantly more Caucasians (p = 0.011) and patients covered by Medicare only (p=0.015) were initiated on pens, whereas significantly more women (p=0.003), Black/African Americans (p=0.011), and dual eligible patients (Medicare + Medicaid; p=0.008) were initiated on vials. Patients initiated on vials had higher median baseline HbA1c values (8.7 vs. 8.4, p<0.001).Survey findings suggested that physicians prescribed vials primarily due to patients' economic constraints (62.9% of vial users vs. only 2.4% for pen users), although a majority of physicians considered pens better than vials (89.2% in terms of adherence, 65.1% in terms of HbA1c control, and 55.4% in terms of resource utilization). Results from this retrospective chart extraction survey suggest that patient characteristics differed between patients initiated on pens vs. vials, and that despite insulin pens being perceived as having better outcomes by physicians, economic considerations play a dominant role in the choice of insulin vials.

INTRODUCTION:While previous studies have noted that hypogonadism (HG) may pose a significant economic and quality-of-life burden, no studies have evaluated the impact of HG on healthcare utilization and costs in the United States.AIM:Compare direct (health care) and indirect (disability leave or medical absence) costs between privately insured U.S. employees with HG and controls without HG.METHODS:The study sample included 4,269 male employees, ages 35-64, with ≥ 2 HG diagnoses (International Classification of Diseases, Ninth Revision, Clinical Modification: 257.2x) or ≥ 1 HG diagnosis and ≥ 1 claim for testosterone therapy, 1/1/2005-3/31/2009, identified from a large, private insurance administrative database that includes medical, prescription drug, and disability claims data. The index date was the most recent HG diagnosis that had continuous eligibility for at least 1 year before (baseline period) and 1 year after (study period). Employees with HG were matched 1:1 on age, region, salaried vs. nonsalaried employment status, and index year to controls without HG.MAIN OUTCOME MEASURES:Descriptive analyses compared demographic characteristics, comorbidities, resource utilization, direct and indirect costs inflated to USD 2009. Multivariate analyses adjusting for baseline characteristics were used to estimate risk-adjusted costs.RESULTS:HG employees and controls had a mean age of 51 years. HG employees compared with controls had higher baseline comorbidity rates, including hyperlipidemia (50.2% vs. 25.3%), hypertension (37.7% vs. 21.1%), back/neck pain (32.0% vs. 15.7%), and human immunodeficiency virus/acquired immunodeficiency syndrome (7.1% vs. 0.3%) (all P < 0.0001). HG employees had higher mean study period direct ($10,914 vs. $3,823) and indirect costs ($3,204 vs. $1,450); HG-related direct costs were $832 (all P < 0.0001). Risk-adjusted direct ($9,291 vs. $5,248) and indirect ($2,729 vs. $1,840) costs were also higher for HG employees (all P < 0.0001).CONCLUSIONS:Employees with HG had higher comorbidity rates and costs compared with controls. Given the low HG-related costs, a primary driver of costs among HG patients appears to be their comorbidity burden.

Compare direct and indirect (workloss) costs between privately-insured U.S. employees with hypogonadism (HG) and demographically matched controls without HG. Male employees, ages 35-64, with ≥2 HG diagnoses (ICD-9-CM: 257.2x) or ≥1 HG diagnosis and ≥1 claim for testosterone therapy between 1/1/2005-3/31/2009 were identified from a privately-insured claims database (N∼12,000,000). The index date was defined as the most recent HG diagnosis with continuous eligibility ≥1 year before (baseline period) and 1 year after (study period). Employees with HG were matched 1:1 on age, region, employment status, and index year to controls without HG. Descriptive analyses compared demographic characteristics, comorbidities, resource utilization, direct costs (reimbursements to providers for medical services and prescription drugs) and indirect costs (disability and medically-related absenteeism) inflated to $2009. Multivariate analyses adjusting for baseline patient differences were used to estimate risk-adjusted costs. 4,269 HG employees, mean age 51, with matched controls met inclusion criteria. Compared with controls, HG employees had higher baseline comorbidity rates: hyperlipidemia (50.2% vs. 25.3%), hypertension (37.7% vs. 21.1%), back/neck pain (32.0% vs. 15.7%), and HIV/AIDS (7.1% vs. 0.3%) (all p<0.0001). HG employees had higher study period rates of inpatient stays (10.8% vs. 5.2%), Emergency Department visits (27.5% vs. 16.3%), outpatient visits (100.0% vs. 76.7%), prescription medication use (95.7% vs. 68.3%), and higher mean workloss days (19.3 vs. 8.8) (all p<0.0001). HG employees compared with controls had higher mean study period direct ($10,914 vs. $3,823) and indirect costs ($3,204 vs. $1,450); HG-related direct costs were $832. HG employees' costs remained higher after adjusting for baseline differences (direct: $9,291 vs. $5,248; indirect: $2,729 vs. $1,840) (all p<0.0001). Employees with HG had higher comorbidity rates and costs compared with demographically matched controls. Given the low HG-related costs, the main driver of overall costs among HG patients may be their comorbidity burden.

To compare real-world healthcare costs after initiation of second-line antidepressant treatment with duloxetine vs. generic selective serotonin reuptake inhibitors (SSRIs) for major depressive disorder (MDD). Study sample was selected from a de-identified U.S. privately-insured claims database (2005-2009). Selection criteria: age 18-64 years, ≥1 MDD diagnosis (ICD-9-CM: 296.2, 296.3) in an inpatient setting or ≥2 MDD diagnoses in emergency room/outpatient settings, initiation of an antidepressant treatment after a 3-month washout, initiation of a second-line treatment with duloxetine or generic SSRI (defined as the index date) no later than 90 days after the end of first-line treatment, continuous coverage eligibility from 6 months before the index date (baseline period) to 12 months after the index date (study period). Using optimal matching on propensity scores and baseline costs, 838 patients initiating second-line duloxetine were matched to 838 patients initiating second-line generic SSRI. McNemar tests were used to compare proportions. Bias-corrected bootstrapping was used to compare healthcare costs (reimbursement to providers for medical services and prescription drugs) inflated to 2009 dollars. Before matching, second-line duloxetine compared with generic SSRI users were older, more likely to be female, had a higher rate of chronic pain, and had higher baseline prescription drug costs and $2,967 higher study period healthcare costs. After matching, there were no notable differences in baseline characteristics and costs. After matching, neither study period healthcare costs ($11,342 vs. $10,328, p=.2316) nor medical costs ($8,377 vs. $8,051 p=.7063) were significantly different between second-line duloxetine and generic SSRI users. Prescription drug costs were significantly higher for second-line duloxetine users ($2,965 vs. $2,277, p<.0001), largely due to differences in mental health-related drug costs ($1,605 vs. $1,008, p<.0001). Controlling for patient differences, this study found that second-line duloxetine users had similar medical costs and higher drug costs compared with second-line generic SSRI users.

96,144 patients initiating sildenafil switched to another PDE-5 agent (57% tadalafil). Prior to switching, the majority of patients (64%) had only one prescription for sildenafil and only 4% of those initiated on a lower dose titrated to the 100mg dose. Mean time to switch from last sildenafil dose was 67–77 days. After switching, about 38% and 34% of tadalafil and vardenafil patients refilled therapy, respectively. Between 22% of tadalafil and 27% of vardenafil patients switched back to sildenafil. CONCLUSIONS: The majority of patients who initiated sildenafil did not switch to a competitor. Among those who did, most either switched back to sildenafil or did not refill treatment.

Many studies suggest that surgery for low back pain (LBP) is excessive, expensive, and often ineffective. We examine the interaction among LBP patients, of demographics, LBP diagnosis categories, co-morbidities, noninvasive therapy use, and medication use to identify diagnostic-treatment pathways that may modify likelihood of LBP surgery using CHAID (Chi-square automatic interaction detector analysis.) The 211,551 patients, ages 18-64 years, with >1 LBP diagnosis, as specified by HEDIS, between 2002 and 2006 were identified from a large administrative claims data. Patients had continuous eligibility >12 months after their index LBP diagnosis (study period) and >6 months before their index diagnosis (baseline period) and no other LBP diagnosis during the baseline period. Exploratory CHAID and logistic analyses were used to identify subgroups of the population most likely to undergo surgery. There were 4,331 patients (2.05%) who had back surgery during the study period (median: 90 days after LBP diagnosis). The likelihood of surgery increased with: age; in men; with severe LBP diagnoses such as spinal stenosis, herniated disc; chronic duration (>3 months); and prior depression. Other co-morbidities (Charlson Index, osteoporosis, asthma) decreased the likelihood of surgery. Use of opiates and corticosteroids appeared associated with increased risk and noninvasive therapies, such as chiropractic and exercise therapies, NSAIDs, and some antidepressants with lower risk. LBP chronicity and diagnostic severity, with combinations of some treatments such as opioids, corticosteroids, chiropractic therapy, NSAIDs, and antidepressants were associated with variations in subpopulation surgery risk rates from >1% to 24%. In summary, surgery risk varies with more conservative, noninvasive therapy and medication combinations, but causality cannot be inferred. Studies to model treatment selection and confounding are needed to assess these interventional prospects for surgery avoidance. (Research funding provided by Eli Lilly and Company, Indianapolis IN.)

Low back pain (LBP) treatment often entails step-therapy additions of medications (including antidepressants) and noninvasive therapies to remediate pain, improve functioning, and avoid surgery. We examine the role of duloxetine and other treatments in the likelihood of LBP surgery, adjusting for potential selection bias and confounding. The 211,551 patients, ages 18-64 years, with >1 LBP diagnosis, as specified by HEDIS, were identified from a large administrative insured claims database. Patients had continuous eligibility at least 12 months after their index LBP diagnosis (study period) and >6 months before their index diagnosis (baseline period) and no other LBP diagnosis during the baseline period. 4,331 patients (2.05%) had back surgery, and 3,756 (1.78%) patients received duloxetine, after their index LBP diagnosis. Logistic regression was used to develop a propensity score predicting study period duloxetine use using: demographics; baseline comorbidities; resource use; medication use; and direct costs. Patients receiving duloxetine were matched to LBP patients untreated with duloxetine based on propensity score and key confounders (e.g., baseline depression, baseline opioid use). Logistic regressions were conducted to assess the impact of study period treatments on back surgery risk for the full matched sample and the non-depressed subsample. We propensity matched 2,521 duloxetine patients to controls. In general, duloxetine was the last of all treatments (including opiates) initiated prior to surgery. Likelihood of surgery was significantly increased for patients with chronic LBP (<3 months) and severe LBP diagnoses. Three treatments significantly reduced the likelihood of surgery: NSAIDs, chiropractic therapy, and duloxetine. No other antidepressants, anxiolytics, narcotics, relaxants, anti-epileptics, or corticosteroids predicted surgery. Results for treatments were the same in the non-depressed subsample. Duloxetine, NSAIDs, and chiropractic treatments were associated with reduced likelihood of LBP surgery. (Research funding provided by Eli Lilly and Company.)

Treatment guidelines suggest that most low back pain (LBP) resolves within a few weeks and that immediate use of imaging and aggressive therapies should be avoided. We assess the actual practice patterns of imaging, noninvasive therapy, medication, and surgery in LBP patients and compare their costs to those of matched non-LBP controls. The 211,551 patients, ages 18-64 years, with >1 LBP HEDIS diagnosis, between 2002 and 2006, were identified from a large administrative claims database. Patients were required to have continuous eligibility >12 months after their index LBP diagnosis (study period) and >6 months before their index diagnosis (baseline period) and no other LBP diagnosis during the baseline period. LBP patients were matched to a random cohort of non-LBP patients by age, gender, and employment status. Univariate analyses described treatment patterns and compared baseline characteristics and 12-month costs. Patients with LBP had significantly higher rates of baseline comorbidities and resource use compared with controls. 47.7% of LBP patients had imaging soon after LBP diagnosis (mean: 34 days; median: 0 days). The majority of LBP patients (69.8%) used medications (mean: 52 days; median:8 days). Opioids were most commonly prescribed (42.1%). 2.05% of patients with LBP had surgery during the study period (mean: 122 days; median: 90 days) after LBP diagnosis. LBP patients were likely to have chiropractic therapy first, followed by muscle relaxants and NSAIDs. LBP patients had significantly higher direct costs compared with controls ($7,211 vs. $2,382, respectively; P < .001), with surgery patients contributing $33,931 in direct costs. Contrary to clinical guidelines, many patients with LBP start incurring significant resource use and as a result, incur expense soon after index diagnosis. LBP is associated with a significant and early cost burden. (Research funding provided by Eli Lilly.) Treatment guidelines suggest that most low back pain (LBP) resolves within a few weeks and that immediate use of imaging and aggressive therapies should be avoided. We assess the actual practice patterns of imaging, noninvasive therapy, medication, and surgery in LBP patients and compare their costs to those of matched non-LBP controls. The 211,551 patients, ages 18-64 years, with >1 LBP HEDIS diagnosis, between 2002 and 2006, were identified from a large administrative claims database. Patients were required to have continuous eligibility >12 months after their index LBP diagnosis (study period) and >6 months before their index diagnosis (baseline period) and no other LBP diagnosis during the baseline period. LBP patients were matched to a random cohort of non-LBP patients by age, gender, and employment status. Univariate analyses described treatment patterns and compared baseline characteristics and 12-month costs. Patients with LBP had significantly higher rates of baseline comorbidities and resource use compared with controls. 47.7% of LBP patients had imaging soon after LBP diagnosis (mean: 34 days; median: 0 days). The majority of LBP patients (69.8%) used medications (mean: 52 days; median:8 days). Opioids were most commonly prescribed (42.1%). 2.05% of patients with LBP had surgery during the study period (mean: 122 days; median: 90 days) after LBP diagnosis. LBP patients were likely to have chiropractic therapy first, followed by muscle relaxants and NSAIDs. LBP patients had significantly higher direct costs compared with controls ($7,211 vs. $2,382, respectively; P < .001), with surgery patients contributing $33,931 in direct costs. Contrary to clinical guidelines, many patients with LBP start incurring significant resource use and as a result, incur expense soon after index diagnosis. LBP is associated with a significant and early cost burden. (Research funding provided by Eli Lilly.)