The number of individuals aged 65 years and older in the United States is projected to increase from 54 million in 2014 to 80 million by 2050.1 With this increase, there has been a corresponding rise in the prevalence of degenerative spinal disorders.2 This is expected to lead to increased demand for spine surgery in the coming years.2 Posterior instrumentation is used in the management of a variety of spinal disorders, including degenerative disk disease, spondylolisthesis, deformity, and spinal trauma.3 Advanced spinal fusion techniques, such as pedicle screw fixation and interbody fusions, have further contributed to the rising rates of posterior instrumented procedures.3–5 In a report published in 2021, spinal fusion accounted for the highest aggregate hospital costs ($14.1 billion across 455,500 procedures in 2018) of any surgical procedure performed in US hospitals.6 The increased utilization of lumbar spinal fusion procedure in the United States is drawing interest from a variety of stakeholders, including patients, physicians, payers, and policymakers.

Several studies have attempted to measure past trends in the utilization and cost of spinal procedures.7–10 Rajaee et al.7 reported that the annual number of spinal fusion discharges increased 137% between 1998 and 2008 in the United States. By contrast, during the same period, laminectomy, hip replacement, and knee arthroplasty yielded smaller increases of 11.3%, 49.1%, and 126.8%, respectively.7 In another study based on the National Inpatient Sample, Sheikh et al.11 reported an 88% increase in spinal fusion procedures from 1998 to 2014. Despite recent works that sought to characterize past trends in orthopaedic spinal procedures, there is a paucity of research that projects future volume and associated costs. There are only two studies that have projected utilization of spinal fusions, which stands in stark contrast to the large bodies of work that project utilization of other orthopaedic procedures, such as total joint arthroplasty.12,13 Heck et al.9 reported that the utilization of posterior spinal fusions is projected to increase between 82% and 102% in 2060, with women and patients aged older than 75 years most likely to see increases in rate of fusion. Neifert et al.8 further projected increases to 2040 in anterior and posterior cervical decompression and fusion of 13.3% and 19.3%, respectively.

Despite the frequent citation of past trends in healthcare policy formulation,12 there is no available literature on future spinal instrumentation volume and associated costs,11 which necessitates the development of projections based on national data. This study aimed to project single-level and multilevel spinal instrumentation procedure volume and associated costs until 2050. We used an alternative model based on past utilization data from the Centers for Medicare & Medicaid Services (CMS) Medicare/Medicaid Part B National Summary12 and a novel time-series forecasting methodology, Prophet.14 We hypothesized that the demand for both single-level and multilevel spinal instrumentation in the United States will increase substantially over the next three decades.

Methods CMS Medicare/Medicaid Data and CPT Codes

This study used data acquired from the CMS Medicare/Medicaid Part B National Summary in patients aged 65 years or older.12 Data included annual procedure counts from 2000 to 2019 and the combined procedural terminology (CPT) codes that were used to identify whether the procedure used spinal instrumentation.12 CPT codes were divided into four categories: single-level, two-four–level, five-seven–level, and eight-twelve–level. CPT codes were also identified for total shoulder arthroplasty, total hip arthroplasty, and total knee arthroplasty to establish a baseline for comparing trends in mean inflation-adjusted reimbursement for spinal instrumentation procedures. The regressions were formed using annual data from the CMS and included data on Medicare Fee-for-Service (FFS) beneficiaries. As each instrumentation category was likely to exhibit a unique trend over time, this study modeled them separately. Data from 2020 were excluded because of COVID-19–related confounding.

Volume and Cost Adjustments

Procedure counts only included FFS patients. To account for this, we normalized the procedure counts using a ratio of FFS to Medicare Advantage patients provided by the Kaiser Family Foundation, as previously reported in the literature.12 We adjusted the total procedure counts for each CPT code using the proportions reported in Table 1. In addition to the primary analysis, we adjusted the volume counts by CMS population data on Medicare beneficiaries. This allowed us to consider potential confounding factors such as increased spinal trends that may have been influenced by changes in Medicare beneficiaries over time. To compare costs across different time periods, we adjusted the costs for inflation using the US Bureau of Labor Statistics' 2019 Consumer Price Index as a reference.15 Finally, we used annual primary total procedure counts for spinal instrumentation volume to generate Prophet time series forecasts between 2020 and 2050.14

Table 1 - Spinal Instrumentation Volume Between 2000 and 2019, Adjusted to Include Medicare Advantage Patients Year Proportion of Medicare Advantage Patients

(r)

Adjusted Volume and Level

(Volume)[1–r]

Single APC Two-Four APC Five-Seven APC Eight-Twelve APC 2000 17% 7,342 — 20,319 — 1,218 — 193 — 2001 15% 10,431 42.07% 22,864 12.53% 1,559 28.00% 195 0.01% 2002 14% 12,907 23.74% 26,163 14.43% 1,558 −0.04% 223 14.32% 2003 13% 15,441 19.64% 29,633 13.26% 1,970 26.44% 217 −2.69% 2004 13% 18,493 19.76% 32,855 10.87% 2,224 12.89% 224 12.17% 2005 13% 21,215 14.72% 36,113 9.91% 2,662 19.69% 330 35.38% 2006 16% 23,925 12.77% 38,360 6.22% 3,168 19.00% 430 30.28% 2007 19% 26,133 9.23% 40,993 6.86% 3,793 19.72% 431 0.26% 2008 22% 31,251 19.58% 45,737 11.57% 4,269 12.57% 483 12.18% 2009 23% 36,238 15.96% 51,025 11.56% 4,904 14.87% 579 19.84% 2010 24% 40,303 11.22% 53,992 5.68% 5,541 13.00% 692 19.49% 2011 25% 42,763 6.10% 56,231 4.28% 6,089 9.90% 701 1.33% 2012 26% 45,631 6.71% 59,504 5.82% 7,043 15.67% 869 23.90% 2013 28% 49,371 8.20% 62,742 5.44% 7,962 13.05% 1,085 24.84% 2014 30% 54,670 10.73% 67,110 6.96% 9,113 14.45% 1,289 18.80% 2015 31% 57,172 4.58% 70,936 5.70% 9,819 7.75% 1,255 −2.60% 2016 31% 59,972 4.90% 75,946 7.06% 10,672 8.69% 1,378 9.82% 2017 33% 59,469 −0.84% 79,190 4.27% 11,685 9.49% 1,469 6.56% 2018 35% 62,608 5.28% 83,832 5.86% 13,380 14.51% 1,492 1.61% 2019 36% 64,350 2.78% 87,253 4.08% 14,000 4.63% 1,620 8.58%

APC = annual percentage change

Prophet Time-series Modeling

We used the time-series analysis method, Prophet, developed by Taylor et al.,14 to predict the procedural counts and costs of single-level versus multilevel spinal instrumentation until the year 2050. For a detailed mathematical description, refer to the original paper.14 The Prophet model is a state-of-the-art time-series algorithm created by Facebook that has been shown to generate reliable and high-quality forecasts in a variety of healthcare settings.16,17 To ensure an appropriate modeling approach, we applied the default setting for the number of change point parameters in the Prophet model. This involves Prophet identifying 25 potential change points, which were uniformly distributed within the first 80% of the accessible data.14 In cases where the data set is relatively small, Prophet automatically adjusts the number of change points to effectively capture notable changes in the time-series trend.

Each Prophet model used a Bayesian framework that relied on past distributions of the parameters for posterior inference to address the uncertainties associated with the projections.14 The Prophet model substitutes τ with a variance inferred from the data, and the model simulates future rate adjustments that mirror those observed in the past. The parameter τ directly influences the model's flexibility in modifying its rate. As τ increases, the model becomes more flexible in fitting the historical data, resulting in a reduction in training error.14 Point forecasts were generated, and the width of the forecast intervals (FIs) for all models was set to 95% to account for the uncertainties in the projections. To mitigate the potential effect of higher variance associated with smaller sample sizes, we used bootstrapping by maintaining 1,000 uncertainty samples.14 This resampling technique allowed us to simulate a larger number of samples from the available data, thereby facilitating more robust estimates of uncertainty in the FIs.

Model Performance, Sensitivity Analyses, and Statistical Methods

To assess the overall goodness of fit of the Prophet time-series models, the period was partitioned into training (first 15 years) and validation data sets (last 5 years), and the value of the normalized root mean square error (NRMSE) was calculated. NRMSE is a widely used measure of the lack of fit between model and data.18–21 The Prophet model projections were compared with alternative model specifications: generalized linear model methods in the form of Poisson and negative binomial regressions, ordinary least squares linear and log-linear regressions, and autoregressive integrated moving average, all of which have been used extensively in the literature to project Orthopaedic procedure volume (Supplemental Table 1,https://links.lww.com/JG9/A337).12,13,21–23 Sensitivity analyses were conducted, and past and future trends were characterized through compound annual growth rates (CAGRs) for both the Prophet and alternative models (Supplemental Tables 2,https://links.lww.com/JG9/A338 and 3, https://links.lww.com/JG9/A339).12,13,21–23 Analysis of variance (ANOVA) tests were conducted to compare the trend components among the four Prophet models (single-level, two-four–level, five-seven–level, and eight-twelve–level) for volume and cost, respectively, and to characterize past (2000 to 2019) and future (2020 to 2050) trends. Statistical analysis was done with the use of the R programming environment (version 4.6.5; R Core Team 2023).

Results

In 2019, the most recent year for which Medicare Part B procedural data were included, a total of 29,072 spinal instrumentation procedures were done nationally in the United States among Medicare patients aged older than 65 years, with associated costs of $33,976,785. Between 2000 and 2019, the prevalence and cost of single-level and multilevel spinal instrumentation procedures all increased substantially (P < 0.0001). Analysis of variance tests indicated a statistically significant difference in the trend components among the four volume models (F (3, 200) = 104.2, P < 0.0001) and the four cost models (F (3, 200) = 67.46, P < 0.0001).

From 2000 to 2019, the CAGRs for spinal instrumentation volume and cost were 11.61% and 9.31% for single-level, 7.64% and 5.67% for two-four–level, 13.18% and 10.89% for five-seven–level, and 14.83% and 11.14% for eight-twelve–level, respectively. From 2020 to 2050, the CAGRs for spinal instrumentation volume and cost are projected to change as follows: 2.04% and 1.86% for single-level, 2.78% and 2.32% for two-four–level, 3.72% and 3.47% for five-seven–level, and 3.40% and 3.29% for eight-twelve–level, respectively. The projected CAGRs for all of the Prophet models indicated that the growth rate for spinal instrumentation volume is expected to increase more markedly compared with costs. This trend suggests that the demand and utilization of spinal instrumentation are anticipated to rise at a faster pace than the corresponding costs associated with the procedures across all levels, from single-level to eight-twelve–level surgeries.

The overall goodness of fit of the Prophet time-series models on the validation data set was measured by the NRMSE and averaged 0.0856 for spinal instrumentation volume (range, 0.0248 for two-four–level spinal instrumentation to 0.1189 for single-level procedures) and 0.0681 for spinal instrumentation cost (range, 0.0117 for two-four–level spinal instrumentation to 0.1189 for single-level procedures), indicating good-to-excellent quality of fits (Supplemental Table 1,https://links.lww.com/JG9/A337). The Prophet, autoregressive integrated moving average, and ordinary least squares-linear models all achieved lower NRMSE compared with ordinary least squares log-linear, generalized linear model Poisson and negative binomial models, and projected more conservative CAGRs.

Tables 1 and 2 summarize spinal instrumentation volume and costs, respectively, between 2000 and 2019 after adjusting for Medicare advantage patients and inflation. Between 2000 and 2019, the adjusted annual spinal instrumentation volume increased by 776% (from 7,342 to 64,350 cases) for single level, by 329% (from 20,319 to 87,253 cases) for two-four levels, by 1049% (from 1,218 to 14,000 cases) for five-seven levels, and by 739% (from 193 to 1,620 cases) for eight-twelve levels. The adjusted cost of spinal instrumentation procedures increased by 502% (from $8,429,736 to $50,747,629) for single-level, by 206% (from $22,704,431 to $69,404,647) for two-four levels, by 670% (from $1,545,433 to $11,893,724) for five-seven levels, and by 459% (from $297,185 to $1,662,171) for eight-twelve levels.

Table 2 - Spinal Instrumentation Cost Between 2000 and 2019, Adjusted to Include Medicare Advantage Patients and the Cost of Inflation Year Proportion of Medicare Advantage Patients

(r)

Consumer Price Index

(c)

Adjusted Cost and Level

(Cost*[1+C])[1–r]

Single APC Two-Four APC Five-Seven APC Eight-Twelve APC 2000 17% +53% $8,429,736 — $22,704,431 — $1,545,433 — $297,185 — 2001 15% +47% $12,550,457 48.89% $26,650,778 17.38% $2,056,502 33.07% $314,528 5.84% 2002 14% +46% $13,658,468 8.83% $28,634,258 7.44% $1,844,009 10.33% $325,604 3.52% 2003 13% +42% $16,487,373 20.71% $31,530,017 10.11% $2,240,882 21.52% $314,471 −3.42% 2004 13% +39% $20,075,344 21.76% $36,006,166 14.20% $2,558,450 14.17% $349,275 11.07% 2005 13% +35% $22,974,739 14.44% $39,334,947 9.25% $3,067,441 19.89% $479,887 37.40% 2006 16% +30% $25,031,776 8.95% $40,526,977 3.03% $3,543,012 15.50% $592,202 23.40% 2007 19% +27% $24,642,085 −1.56% $39,132,449 −3.44% $3,824,296 7.94% $524,361 −11.46 2008 22% +22% $27,563,001 11.85% $40,664,869 3.92% $4,007,477 4.79% $555,067 5.86% 2009 23% +22% $32,406,942 17.57% $45,996,129 13.11% $4,674,692 16.65% $667,899 20.33% 2010 24% +19% $35,845,504 10.61% $48,382,232 5.19% $5,241,612 12.13% $771,751 15.55% 2011 25% +17% $38,524,605 7.47% $51,064,083 5.54% $5,825,555 11.14% $796,631 3.22% 2012 26% +14% $39,447,998 2.40% $51,776,749 1.40% $6,462,321 10.93% $939,380 17.92% 2013 28% +12% $41,555,435 5.34% $53,141,865 2.64% $7,071,442 9.43% $1,116,147 18.82% 2014 30% +10% $46,458,031 11.80% $57,290,168 7.81% $8,206,592 16.05% $1,361,269 21.96% 2015 31% +9% $48,826,975 5.06% $60,943,729 6.38% $8,936,888 8.90% $1,342,583 −1.37% 2016 31% +6% $49,453,811 1.28% $63,161,897 3.64% $9,447,866 5.72% $1,462,652 8.94% 2017 33% +4% $48,255,815 −2.42% $64,876,563 2.71% $10,222,941 8.20% $1,544,224 5.58% 2018 35% +2% $49,940,944 3.49% $67,455,460 3.98% $11,486,343 12.36% $1,550,829 0.43% 2019 36% — $50,747,629 1.62% $69,404,647 2.89% $11,893,724 3.55% $1,662,171 7.18%

APC = annual percentage change

Tables 3 and 4 summarize spinal instrumentation volume and cost projections between 2020 and 2050, after adjusting for Medicare advantage patients and inflation. From 2000 to 2019, the mean inflation-adjusted reimbursement for single-level spinal instrumentation procedures decreased 45.6% from $1,148.15 to $788.62. The mean inflation-adjusted reimbursement for other prevalent orthopaedic procedures decreased as follows: total shoulder arthroplasty $1,941.46 to 1,492.19 (−23.1%), total hip arthroplasty $2,312.01 to 1,406.14 (−39.2%), and total knee arthroplasty $2,439.81 to 1,406.20 (−42.4%). The mean inflation-adjusted reimbursement for single-level spinal instrumentation procedures is projected to decrease an additional 4.2% by 2050, to $756.89. By 2050, the number of single-level spinal instrumentation procedures performed yearly is projected to be 124,061 (95% FI, 87,027 to 142,907, Figure 1), with an associated cost of $93,900,672 (95% FI, $80,281,788 to $108,220,932), and the number of two-four level spinal instrumentation procedures performed yearly is projected to be 214,517 (95% FI, 191,425 to 237,425), with an associated cost of $146,323,437 (95% FI, $135,874,961 to $157,427,669). Figure 2 shows volume trends from 2000 to 2019, after adjusting for the changing Medicare population. Figure 2 is not meant to present the true national volume counts of spinal instrumentation procedures but to more accurately assess trends in utilization after adjusting for Medicare beneficiaries as a potential cofounder. Notably, two-key inflection points can be visualized: a rapid increase in single-level instrumentation beginning in 2008 and a plateau of single-level instrumentation beginning in 2017 (Figure 3).

Table 3 - Prophet, Bayesian Inference, Volume Projections (2020-2050) for Spinal Instrumentation Year (5-yr Intervals) Spinal Instrumentation (No. of Procedures) Single (95% FI) Two-Four (95% FI) Five-Seven (95% FI) Eight-Twelve (95% FI) 2020 66,408 [56,488-56,702] 91,819 [91,127-92,406] 14,965 [14,817-15,121] 1,747 [1,672-1,814] 2025 76,049 [63,994-68,553] 112,464 [110,657-114,347] 20,257 [19,512-20,976] 2,307 [2,214-2,390] 2030 85,691 [70,303-81,667] 132,558 [128,053-137,384] 25,587 [23,796-27,438] 2,815 [2,668-2,956] 2035 95,267 [75,546-95,561] 152,992 [144,967-161,006] 30,790 [27,622-33,934] 3,330 [3,097-3,572] 2040 104,778 [80,435-110,818] 173,782 [161,636-186,468] 35,867 [31,314-40,975] 3,852 [3,508-4,197] 2045 114,419 [83,913-126,754] 194,427 [177,376-212,372] 41,15