Predicting Sepsis Mortality and Costs using Medicare Claims: A Method to the Madness

Sepsis mortality has declined over the last three decades[1–3] against the odds of growing patient age and complexity, an absence of specific diagnostic modalities, and a litany of unsuccessful clinical trials of proposed directed therapeutics. Notwithstanding, sepsis remains a colossal burden on our healthcare system in its occurrence and associated morbidity, mortality and costs[3, 4] From the standpoint of awareness and large-scale action, the field of sepsis has greatly benefited from public funding, advocacy groups, policy makers, economists, and payers having a seat at the table that until recently seated predominantly healthcare professionals. However, any action by these entities must be driven by high-quality data, which sits on a bedrock of sound analyses. Advanced modelling techniques are increasingly utilized to analyze risk, costs, historical trends and direct avenues of future research, and familiarity with these methodologies could harmonize interpretations across the gamut of sepsis stakeholders, optimize resource allocation, and ultimately improve patients’ outcomes. Buchman and colleagues have undertaken an ambitious and comprehensive analysis of the burden of sepsis in the Medicare beneficiary population as a joint venture between the Biomedical Advanced Research and Development Authority (BARDA) and Centers for Medicare and Medicaid Services (CMS) to provide detailed assessment of the morbidity and costs of this exceedingly prevalent condition in an ever-growing and vulnerable population. Armed with these data, providers and researchers may define new modifiable risk factors to facilitate both improvements in patient care as well as public health interventions and public policy.

In this third report in the series[5], Buchman et al. sought to describe for readers the methodology of data analysis used in the first two reports[6, 7]. Additionally, they used Medicare claims data in conjunction with the Center for Medicare and Medicaid Service Hierarchical Condition Codes (CMS-HCC) risk adjustment model (used primarily by CMS to determine reimbursement) to evaluate the impact of sepsis, its severity and time of onset, as well as patient age, and key comorbidities on short and long-term mortality risk following an acute inpatient admission. They then created informative models to provide a contemporary lower-bound estimate of tangible costs of sepsis care. For health-outcome modelling, the study population was restricted to Medicare Fee-For-Service (FFS) admissions in 2017 in order to circumvent the ICD-9-to-ICD-10 transition period. Notably, only beneficiaries’ first sepsis admission in 2017 was counted and inclusion required no prior inpatient sepsis admission in the year prior to admission. For financial forecasting, data from the Medicare FFS and SNF claims between 2012 and 2018 were used to train linear regression models, which were then used to project 12-month and monthly cost estimates, with and without a “seasonality” factor to ultimately forecast 2019 estimates. As evidenced by their discovery of a blip in sepsis costs likely attributable to hypervirulent influenza in winter of 2017–2018[8], authors showcase this type of “near-real-time” analysis of claims data as a potential means to target public health measures, such as those that could prevent further transmission in an outbreak scenario.

While it may be intuitive that patients with more comorbid conditions are at a higher risk of death from sepsis, this analysis provides quantitative data on a number of pre-existing conditions of particular clinical and economic interest. The mortality estimates provided in this report are striking. Compared to inpatient encounters without sepsis, average predicted probability of death within one week of inpatient discharge was notably 4-fold higher for sepsis admissions and more than 9-fold higher for those with septic shock. For patients with a documented history of malignancy, the probabilities of death were even higher and similar to a recent report which cited increasing mortality proportional to the number of dysfunctional organ systems[9]. Buchman et al. data also support previously described worsened outcomes for hospital-onset versus community-onset sepsis[10]; nosocomial sepsis complicating an abdominal surgery carried more than a 11-fold risk compared to abdominal sepsis-on-admission warranting surgical source control, underscoring the importance of acknowledging event sequence (sepsis, surgery) on survival and potentially on re-imbursement. Six-month mortality estimates followed similar patterns. However, three issues should be noted here: in clinical practice, the behavior of multiple comorbidities is often synergistic rather than simply additive, which cannot be captured using an approach that did not account for potential interactions; second, the odds ratios from the influencers were based on univariate analyses; and third, comorbidities are known to be highly associated with early but not with late sepsis mortality. A recent study[11] found that most deaths immediately following sepsis were not preventable and were tied closely to multiple underlying illnesses. However, if patients survive the immediate post-sepsis, the incremental impact of comorbidities on their sepsis-associated longer-term survival was small, as has been described using a more granular dataset of Medicare beneficiaries that participated in the Health and Retirement Study[12]. While some pre-existing conditions may remain inconsequential in the long-term, others will remain increasingly difficult-to-resolve such as incurable cancer. Hence, focusing on what we know works to improve long-term outcome after sepsis: earlier mobilization and simple remedies to prevent repeated hospitalizations due to heart-failure, COPD exacerbations, renal failure, recurrent urinary tract infections and aspiration all should be prioritized by stakeholders[13].

Buchman et al. found that on average the inpatient cost of sepsis continued to rise by about 5% per year between 2012 and 2018, surpassing inflation rates. When combining Medicare Advantage (MA) with FFS, the inpatient and Skilled Nursing Facility (SNF) cost of sepsis care for Medicare beneficiaries in 2019 was projected at $45 billion. After applying National Inpatient Sample estimates [14] as a correction factor to account for the entire US population as well as increasing Medicare enrollments and private sector and SNF patients, authors suggested a lower-bound, rough-order-of-magnitude estimate of $62 billion as the annual cost of sepsis care in the US. While these astronomical cost projections may be of limited utility to individual providers, they have relevant implications for resource allocation for public health initiatives or implementation science and hence warrant close scrutiny. The cost estimate differs considerably from a previous estimate by Torio and Moore[4], in which 2013 inpatient costs of “septicemia” billed specifically to Medicare, derived from the Nationwide Inpatient Sample, totaled $14.6 billion. However, Buchman et al. provide reasonable justification for the observed difference accounting for the rising healthcare costs and relative growth over 6 years in the Medicare eligible population, as well as the incorporation of SNF costs and expanded sensitivity of all explicit sepsis codes (versus septicemia alone). Although Buchman et al. have sound rationale to limit their report to a conservative estimate, it behooves us to explore what the potential realistic ceiling of annual cost estimate of sepsis might be to Medicare or society at large and we have identified three elephants in the room in this regard: (1) due to the low sensitivity of explicit sepsis codes (found to be 32.3% in a large study[3]) two-third of cases are essentially being missed. Even though Medicare reimbursement for these cases is occurring without the sepsis label, they do in fact contribute to the cost of (unlabeled) sepsis. Extrapolating this level of sensitivity to the lower bound rough-order-of-magnitude estimate provided here, one could estimate up to an additional $40 billion in sepsis care potentially not captured by the explicit coding algorithm. (2) Second, as alluded to an extent by the authors, there remain massive intangible secondary costs that are beyond the scope of this analysis but may have significant impact on the US economy. While sepsis may disproportionately affect the aged with limited remaining quality-adjusted life years, the social and economic ramifications do not end there. As more Medicare beneficiaries survive sepsis admissions and ultimately return home, they may still require significant assistance. According to a 2018 AARP Public Policy Institute report[15], 6.2 million Millennials (born 1980–1996) are now acting as a caregiver to a parent, in-law, or grandparent. Notably, nearly half of these care-givers are men (47%), which is higher than previous generations (39% Generation X and 38% Boomers), and 72% of those caregivers are of Hispanic or African American ethnicity. The effects on productivity of this growing segment of the young population are hard to estimate, but are most certainly real. (3) A final consideration when interpreting these data is that only beneficiaries’ first sepsis admission in the study period was analyzed. Given the exceedingly high readmission rates following sepsis[6, 7], their estimate excludes a large and costly sector of the Medicare population. Indeed, 386,674 (34%) of the total sepsis admissions in 2017 were excluded for this reason which obviously impacts the overall true cost burden and forecast.

Future efforts of this nature could include additional at-risk populations such as patients without claims but with clinical indicators of sepsis, the Veterans Affairs population, patients receiving private-sector SNF care as well as those repeatedly admitted for sepsis, potentially bringing us one step closer to the scientific truth that all of us as stakeholders of sepsis earnestly seek. While the methodology presented by Buchman et al. may be a bit detailed and not readily digestible in a single read, it provides a toolkit for expansion of these data and analyses to other patient populations or prevalent conditions, which could be used to address myriad important clinical questions. With respect to this series of articles on sepsis in an aged population, even though they are not fully nationally representative, they represent a massive 100% sample of continuously enrolled Medicare beneficiaries in the United States, and provide a pathway to better reliability and dissemination of the actual consequences of sepsis to our society.