Can Discharge Radiographs Predict Junctional Complications? A Decision Tree Analysis

Abstract

Study Design

Retrospective cohort study.

Objectives

To determine if standing pre-discharge radiographs can predict the development of junctional complications.

Materials and methods

Adult spinal deformity patients who underwent fusion of the lumbar spine (≥5 levels, LIV pelvis) were included. All patients underwent full-length standing radiographs before hospital discharge. Outcomes of interest included 2-year radiographic PJK and proximal junctional failure (PJF). Patients were stratified into 3 exclusive groups: No PJK, PJK, and PJF. Chi-square automatic interaction detection (CHAID) decision tree analysis was utilized to identify pre-discharge proximal junctional angle (PJA) thresholds associated with increased risk of PJK or PJF.

Results

The 117 study patients had a mean age 65.8 ± 8.5, BMI 27.2 ± 4.9, PI-LL 23.3 ± 17.4, TPA 27.2 ± 11.5. Sample was stratified into 64 (54.7%) No PJK, 39 (33.3%) PJK, 14 (12.0%) PJF. No differences were detected between cohorts in discharge alignment, preop-discharge change, or offset from age-adjusted alignment targets (P > .005). Decision tree analysis showed that the first branch point depended on the UIV, as most patients with an UT UIV did not develop PJK or PJF (no PJK, 67.4%). For patients with an LT UIV, a second branch point occurred based on the ΔPJA. 89.5% of LT patients with a ΔPJA < 4.3° were free of radiographic PJK and PJF. The third branch point occurred based on the PJA at discharge. Thus, the highest risk group was comprised of ΔPJA ≥4.3° and PJA > 15.5°, as 57.1% of developed PJF and 28.6% PJK.

Conclusion

Most patients with a lower thoracic UIV, preop-discharge ΔPJA ≥4.3°, and discharge PJA > 15.5° develop PJF.

Introduction

Proximal junctional kyphosis (PJK) is the most common mechanical complication after adult deformity surgery (ASD), occurring at a rate of approximately 20-50%, depending on the series.^1-3 Notably, proximal junctional complications present along a spectrum, from an asymptomatic increase in the proximal junctional angle (PJA) to mechanical failure necessitating revision. ⁴ The drivers of proximal junctional complications are multifactorial, consisting of ligamentous, bony, and implant complications.^5-7 Long-term studies show that once recognized, proximal junctional complications tend to progress, which can eventually impact quality of life.^2,8 This progression makes sense from a physiologic standpoint—as PJK increases the anterior sagittal alignment, the body’s compensatory mechanisms (pelvic retroversion) engage. In patients who are fused to the pelvis, this creates a vicious cycle by angulating the (instrumented or fused) spine posteriorly, further increasing the load at the proximal junction.^9,10 Thus, our ability to identify patients at the beginning of the “proximal junctional cascade” is essential.

Many authors have attempted to calculate the ideal surgical correction needed to prevent long term PJK.^11,12 Notably, these studies are predicated on using the 2-8 week outpatient office XR as the “initial postoperative alignment,” which is then used to calculate and compare surgical corrections. ¹³ However, between 65-79% of radiographic PJK is already present by the first postoperative visit.^2,6 Therefore, the literature to this point has missed a key window of time in which the proximal junctional cascade begins. By obtaining standing radiographs before hospital discharge, we may be able to identify at risk patients earlier in the PJK cycle or identify new corrective goals. The purpose of this study is to determine if standing discharge radiographs can be used to predict the development of PJK after surgery for ASD.

Methods

Patient Sample

This study was approved by the institutional review board. A retrospective review was performed from an institutional database consisting of adult patients (≥18 years old) with radiographic adult deformity (coronal Cobb angle >20°, C7 sagittal vertical axis (SVA) >5 cm, pelvic tilt (PT) >25°, and/or thoracic kyphosis (TK) >60°). The database was screened for patients who underwent ≥5 level fusions to the pelvis. All patients underwent single-stage, posterior only fusion between 2013-2018. Vertebroplasty was not used, tether use was still in an experimental phase during the surgical period and consisted of varying forms in a non-standardized manner. ³⁸ Patients with less than 2 years of follow up, tumor or neuromuscular disease were excluded.

Data Collection

Clinical and surgical data were collected from the electronic medical record. All patients underwent preoperative, pre-discharge, and two-year full-length standing lateral and AP radiographs. Pre-discharge radiographs were performed once the patient could stand unassisted and usually as close as possible to the date of hospital discharge (day before or day of). The following radiographic parameters were measured at all time points: incidence (PI), PT, lumbar lordosis (LL), pelvic-incidence lumbar lordosis mismatch (PI-LL), T2-T12 thoracic kyphosis (TK), C7 sagittal vertical axis (SVA), T1 pelvic angle (TPA) and the proximal junctional angle (PJA), defined as the angle between the caudal endplate of the uppermost instrumented vertebra (UIV) and the cephalad endplate of 2 suprajacent vertebrae above the UIV. UIV was classified as upper or lower thoracic (UT above T8, LT T8 or below). To determine overall deformity correction, offset from age-adjusted postoperative alignment goals were calculated. ¹⁴ Outcomes of interest included two-year PJK (defined as a PJA >10° and a pre- to post-operative kyphotic increase of PJA >10°) and proximal junctional failure (PJF, defined as a proximal junctional angle [PJA]>28° and ΔPJA >22° or revision for PJK before 2 years). ¹⁵

Statistical Analysis

All analyses were performed using IBM SPSS Version 22.0 (Armonk, NY). Significance was defined as a type I error rate of 5% (P < .05). Patients were stratified into 3 mutually exclusive groups based on their 2-year outcomes: No PJK, PJK, and PJF. Categorical variables were compared using chi-square, continuous variables were compared using ANOVA. Chi-square automatic interaction detection (CHAID) decision tree analysis was utilized to identify pre-discharge PJA thresholds that were associated with an increased risk of PJK or PJK. At each step, CHAID chooses the independent (predictor) variable that has the strongest interaction with the dependent variable. The growing parameters for the decision tree analysis included a minimum parent node of 20 and a minimum child node of 10. Finally, a sensitivity analysis was performed using classification and regression decision tree analysis (CRT). Briefly, CRT is like CHAID but uses slightly different methods of node-splitting.^16,17

Results

Patient Sample

A total of 117 patients were identified from the database. The sample was middle-aged and overweight (mean age 65.8±8.5 years, mean body mass index [BMI] 27.2 ± 4.9 kg/m2, and 78% female). Mean baseline alignment parameters showed a PI of 53.0 ± 12.2°, PT of 27.0 ± 9.2°, PI-LL of 23.3 ± 17.4°, SVA of 87.7 ± 73.1 mm, and T1PA of 27.2 ± 11.5°. The average hospital length of stay (LOS) was 6.3 ± 2.7 days. Sagittal alignment parameters were significantly different between preoperative, discharge, and two-year radiographs (P < .001 for all comparisons).

Incidence of and Risk Factors for PJK and PJF

The incidence of PJK at 2 years was 54.7% while the incidence of PJF was 12.0%. For the patients with PJF, revision occurred at a mean of 10.9 ± 10.7 months. Except for a higher rate of revision surgery in the PJK cohort, no differences were detected between the cohorts in demographics, surgical parameters, discharge alignment, preop-to-discharge change in alignment, or offset from age-adjusted alignment targets (Tables 1 and 2, Figure 1). The specific UIV (ie, T11 vs T9 vs T4, etc.) was also not significantly different between the cohorts (P > .05). Subgroup analyses of the cohort in UT and LT subgroups also showed no differences in offset from age-adjusted alignment targets (P > .05 for all comparisons).

Table 1.

Comparison in Demographics, Comorbidities, Surgical Data, and Preoperative Alignment Between PJK, PJF, and No PJK Cohorts.

	No PJK (64, 54.7%)	PJK (39, 33.3%)	PJF (14, 12.0%)	P-value
Age, yrs, mean (SD)	64.5 (9.7)	66.9 (6.4)	69.1 (6.6)	.128
Female, N (%)	52 (82%)	32 (82%)	12 (85%)	.925
BMI, mean (SD)	26.9 (5.2)	27.2 (4.2)	29.0 (5.3)	.359
ASA class, (25th/50th/75th %tile)	2 | 2 | 2	2 | 2 | 2	2 | 2 | 3	.809
Revision Surgery, N (%)	33 (52%)	10 (26%)	7 (50%)	.039
LT UIV, N (%)	31 (48%)	26 (67%)	10 (71%)	.198
Preop sagittal alignment, mean (SD)
PI	53 (12.9)	52.3 (12.5)	55.3 (7.7)	.735
Pelvic tilt	26.7 (9.2)	27.1 (9.8)	28.1 (8.1)	.877
PI-LL	22.7 (17.5)	24.7 (18)	22.1 (16.1)	.825
SVA, mm	84.9 (69.8)	89.5 (80.3)	95.7 (71.3)	.871
T2-T12	−36.9 (22.4)	−33.8 (21.8)	−45.6 (13.4)	.207
T1PA	26.6 (11.1)	27.6 (12.9)	28.4 (9.9)	.847

Table 2.

Comparison of Alignment at Discharge and Change in Pre-Operative to Discharge Alignment Between PJK, PJF, and No PJK Cohorts.

		No PJK (64)		PJK (39)		PJF (14)		P-value
		Mean	SD	Mean	SD	Mean	SD
Discharge Align.	PI	53.0	12.8	52.3	11.7	54.9	8.5	.794
	PT	17.0	9.5	15.3	10.9	18.4	6.6	.529
	LL	52.8	15.5	54.5	9.8	59.0	8.3	.283
	PI-LL	0.1	12.8	−2.2	11.5	−4.1	6.5	.385
	TK	−45.4	17.1	−45.8	13.2	−55.7	11.9	.073
	TPA	12.9	9.4	12.2	9.7	13.2	6.0	.909
	SVA	19.5	42.8	24.4	43.3	11.1	39.7	.598
Change pre-to-dis	PI	0.0	2.0	0.0	2.5	−0.4	1.9	.764
	PT	−9.7	8.2	−11.8	9.0	−9.7	9.4	.449
	LL	22.6	14.7	26.9	16.5	25.8	14.7	.366
	PI-LL	−22.6	14.6	−26.9	16.5	−26.3	13.8	.345
	TK	−8.8	11.1	−12.0	15.1	−10.0	10.5	.458
	TPA	−13.7	9.9	−15.4	9.8	−15.2	8.5	.669
	SVA	−65.7	66.4	−65.1	64.6	−84.6	64.5	.593

Figure 1.

Offset from age-adjusted alignment goals, compared between cohorts. Note that all cohorts were overcorrected. No differences in offset were detected for any correction parameter (P > .05 for all comparisons).

Decision Tree Analysis

Decision tree analysis showed that the first branch point depends on the UIV (Figure 2). Namely, most patients with an UT UIV did not develop PJK or PJF (no PJK, 31, 67.4%, P = .039). For patients with an LT UIV, a second branch point occurred based on the ΔPJA from preoperative to discharge. 89.5% of LT patients with a ΔPJA < 4.3° at discharge were free of radiographic PJK and PJF (P = .001). The third branch point occurred based on the PJA at discharge. For patients with an LT UIV, ΔPJA ≥4.3°, and PJA<15.5°, 55.3% developed radiographic PJK, but only 7.9% developed PJF (P = .010). The highest risk group was comprised of ΔPJA ≥4.3° and PJA > 15.5°, as 57.1% of developed PJF and 28.6% radiographic PJK (P = .010).

Figure 2.

Decision tree analysis of factors predictive of proximal junctional complication at 2 years.

Using this decision tree analysis, an example scenario could be proposed in which a hospital chooses to “treat” (ie, implement early or more frequent follow up, brace, etc.) those patients with an LT UIV, ΔPJA ≥4.3°, and PJA>15.5. If such a strategy was employed, then for patients with LT-pelvis fusions, treatment would be applied to 66.7% (8/12) of the patients who would have gone on to a junctional failure (as per the decision tree) and 6.1% (2/33) patients who would not have developed a junctional complication. In contrast, 4 patients (33.3%) who would have gone on to PJF would not have been treated.

Discussion

Our retrospective review of 117 surgically treated ASD patients at 1 institution is the first to show that there is an association between alignment on discharge radiographs and the development of proximal junctional complications at 2 years. The overall rate of PJK and PJF was 33.3% and 12.0%, respectively, like previous series.^18-20 Notably, many of our patients were overcorrected compared to age-adjusted alignment goals (Figure 1). This occurred because the surgeries were performed between 2013-2018, a time period before the idea of age-adjusted alignment goals was popularized.^21,22 As such, this likely contributed to the high rate of PJK seen in this study. While comparisons of demographics, surgical factors, alignment, and offset from age-adjusted goals showed no differences between the cohorts, our decision-tree analysis revealed that the combination of LT UIV, ΔPJA ≥4.3°, and PJA>15.5 was associated with a high risk of developing PJF. These findings have several implications for postoperative decision-making and spur new questions regarding the use of discharge radiographs.

The initial branch point of our decision tree was based on the UIV. Several previous authors have shown that an LT UIV is associated with the development of PJK.^20,23-25 Daniels et al followed a multicenter sample of 303 patients for 2 years, finding that a UT LIV was associated with half the odds of developing PJK compared to an LIV in the LT region. ²⁶ However, this finding has been contrasted by previous investigations showing no difference in the rate of PJK whether the fusion ends in the LT or UT regions.^25,27 Within the LT spine, the specific UIV level has also been controversial, with investigators debating the risks of ending the construct in the thoracolumbar junction.^28,29 Our investigation may bring some clarity to this controversy, as our analyses showed that the UIV itself was not a factor when analyzed alone, but only when combined with other parameters (PJA and ΔPJA). This seems to suggest that PJA and change in PJA has a different significance based on the UIV, as has been seen in previous attempts to define “consequential” types of junctional angle changes. ³⁰ The UT vs LT debate may not be sufficient to explain the risk of PJK, a more nuanced consideration of other factors is necessary.^9,20,31

The second branch point of the decision tree involved the change in PJA. Interestingly, the change in PJA was about half that of Glattes and colleague’s PJK definition (PJA >10° and a pre- to post-operative kyphotic increase of PJA >10°). ³² In the Glattes series, the authors state that radiographs were taken at “preoperative, early postoperative and most recent follow up”, but do not define when the first postoperative XR was taken. Assuming that the “early postoperative” XR was taken at a similar time point as other series from their institution (ie, 2-8 weeks), ¹⁸ it is fascinating to note that we found a change in PJA of approximate half that of Glattes’s (4.3°) as the critical value when XRs were taken at an earlier time point (ie, at discharge). The importance of avoiding an increase at the PJA postoperatively has also been demonstrated in prior series. ⁸ We showed that for those with an LT UIV, even a slight acute increase in the PJA predisposed the patient to PJK. Namely, 53.5% of these patients developed PJK or PJF, compared to 10.6% in those whose PJA increased less than 4.3°.

The third point of the decision tree is in line with past literature demonstrating that there is a certain “absolute PJA value” at which PJK progresses. Park et al followed 73 patients for over 3 years to evaluate the long-term outcomes of patients with asymptomatic radiographic PJK (defined as a PJA of >10°). ⁸ In their series, the PJK cohort had an initial PJA of 6.5° at their first postoperative radiograph. This discrepancy between ours lies in the difference in patient samples—specifically, they excluded any patients who had PJF or a revision for PJK. In contrast, PJF was an outcome of interest in our study, which may explain why we found that an initial PJA of 15.5° was a risk factor. A PJA of this degree could represent a mechanistic breakpoint in the junctional cascade, at which point the forward sagittal alignment is enough to induce the body’s compensatory pelvic tilt, which induces further posterior orientation of the UIV and higher forces at the junction.^9,31 Future research will be needed to determine whether postoperative interventions (e.g., bracing or physical activity limitations) that counteract these forces could impact the progression to PJK.

The ability to identify patients at high risk for progression of PJK before they leave the hospital has 3 clinical consequences (Figure 3). First, discharge radiographs should be a standard protocol for any patient undergoing ASD surgery. Pre-discharge counseling can help modify patient expectations, helping them to understand that their risk for needing revision surgery is higher than that discussed preoperatively. Maintaining accurate patient expectations is key to satisfaction and building patient-surgeon trust. Second, while postoperative bracing has yet to show any advantage in preventing mechanical complications after ASD surgery, we are unaware of a study that has specifically braced high-risk patients only.^33,34 Our analysis allows for the identification of a high-risk subset of patients very early in the postoperative course. Finally, the decision tree helps us to understand that if the thresholds were to be applied, resources (ie, earlier follow up, bracing, etc.) would be used for appropriate patients. For example, if a hospital chooses to implement early follow up and bracing for all patients with an LT UIV, ΔPJA ≥4.3°, and PJA>15.5, then this would ensure treatment for 66.7% of all patients who would have experienced a junctional failure, while only 6.1% of the patients who were not going to develop a junctional complication would be treated (ie, “over-treated” for a complication they would not have developed anyway). Future analyses will be needed to determine whether any intervention (e.g., revision before hospital discharge, early follow-up, bracing) would change these patients’ ultimate outcome.

Figure 3.

Case example of a 60 year old female who underwent T11-pelvis fusion with an L4 pedicle subtraction osteotomy. Comparison between preoperative (A) and discharge radiographs (B) demonstrated an increase in PJA from 5.0 to 15.8°. At 9 months the patient developed a symptomatic proximal junctional failure (C) which required revision surgery to extend the fusion to T4 (D).

Our study has several limitations. First, we could not account for many factors that have been just recently implicated in PJK. Our decision tree did not account for the bone density at the UIV, the muscle quality, or the patient’s preoperative thoracic flexibility.^6,35,36 For example, while the predominance of females in our cohort was in line with previous series on adult deformity, this may have correlated with a similar predominance of osteoporosis, a known risk factor for PJK.^6,8,37 In another example, the impact of tether use remains unknown. After an analysis of posterior ligamentous reinforcement (PLS) with nylon tape performed on this same institutional dataset (29% PLS in the years 2014-2017), we did not find an association with PLS use (27.3% vs. 28.6%, P = .827). ³⁸ Despite other studies showing promising results, ⁷ we subsequently discontinued the practice. As we understand and incorporate more of these factors into our preoperative assessment, we will be able to better risk-stratify patients on a more individual basis. Second, our analysis was based on a single-institution series. This provides a certain control, including parameters not necessarily quantifiable or collectable (same post-operative protocol, geographic region, similar treatment, team experience, etc.) which help the prediction accuracy but may not be generalizable directly to other institutions. Third, our findings only apply to predictions up to 2 years. PJK can occur 3 or more years after surgery, ⁸ thus our analysis cannot be used to extrapolate outcomes beyond 2 years. Fourth, our sample size is relatively small, thus some non-significant findings (eg, the lack of significance in age differences) may be due to type II error. This sample size also limited the number of predictive factors that can be analyzed at one time. Finally, there are some methodologic limitations of the CHAID decision tree method compared to other decision tree methods. Specifically, CHAID tends to simplify patients to a limited number of nodes. As data (ie, bone density, muscle quality, flexibility) becomes more available and quantifiable, the development of more complex decision trees (ie, neural networks) will be possible. However, these decision algorithms do not allow for analyses at the “nodal level”, making their processes harder to implement in the case of a complex or “deep” decision tree.

Conclusions

This is the first series to show an association between pre-discharge alignment and two-year junctional complications. The majority (85.7%) of patients with lower thoracic UIV, ΔPJA ≥4.3°, and PJA > 15.5° develop proximal junctional complications, with 57.1% developing PJF. Notably, this “risk profile” will need to be validated in a larger, external dataset before we can say that such patients will definitively go on to PJK/PJF. The ability to identify these patients early may allow for earlier postoperative interventions in this high-risk patient cohort, potentially mitigating their risk of future complications. Discharge radiographs should be standardized after adult deformity surgery, which may lead to even more nuanced findings about the relationship between discharge alignment and mechanical complications. As researchers deepen their understanding of the drivers behind PJK, discharge alignment can fall in with the myriad of other factors to consider in developing patient-specific risks and treatments.

ORCID iDs

Francis Lovecchio https://orcid.org/0000-0001-5236-1420

Bryan Ang https://orcid.org/0000-0002-4174-997X

Virginie Lafage https://orcid.org/0000-0002-0119-7111