The Spine Journal 17 (2017) 390–400

Clinical Study

Preoperative estimation of disc herniation recurrence after

microdiscectomy: predictive value of a multivariate model based on
radiographic parameters
Evgenii Belykh, MDa,b,c,1, Alexander V. Krutko, MD, PhDd,1, Evgenii S. Baykov, MD, PhDd,
Morgan B. Giers, PhDa,b, Mark C. Preul, MDa,b, Vadim A. Byvaltsev, MD, PhDa,c,e,*
a
Irkutsk Scientific Center of Surgery and Traumatology, Bortsov Revolyutsii str., 1, Irkutsk, 664003, Russia
b
Department of Neurosurgery, Barrow Neurological Institute, St. Joseph’s Hospital and Medical Center, 350 W. Thomas Rd, Phoenix, AZ 85013, USA
c
Department of Neurosurgery, Irkutsk State Medical University, Krasnogo vosstaniya str., 1, Irkutsk, 664003, Russia
d
Neurosurgery Department No. 2, Novosibirsk Scientific Research Institute of Traumatology and Orthopedics, Frunze str., 17, Novosibirsk, 630091, Russia
e
Neurosurgery Department, Irkutsk Railway Clinical Hospital, Botkina str, 10, Irkutsk, 664005, Russia
Received 14 July 2016; revised 27 September 2016; accepted 12 October 2016

Abstract BACKGROUND CONTEXT: Recurrence of lumbar disc herniation (rLDH) is one of the unfa-
vorable outcomes after microdiscectomy. Prediction of the patient population with increased risk of
rLDH is important because patients may benefit from preventive measures or other surgical options.
PURPOSE: The study assessed preoperative factors associated with rLDH after microdiscectomy
and created a mathematical model for estimation of chances for rLDH.
STUDY DESIGN/SETTING: This is a retrospective case-control study.
PATIENT SAMPLE: The study includes patients who underwent microdiscectomy for LDH.
OUTCOME MEASURES: Lumbar disc herniation recurrence was determined using magnetic res-
onance imaging.
METHODS: The study included 350 patients with LDH and a minimum of 3 years of follow-up.
Patients underwent microdiscectomy for LDH at the L4–L5 and L5–S1 levels from 2008 to 2012.
Patients were divided into two groups to identify predictors of recurrence: those who developed rLDH
(n=50) within 3 years and those who did not develop rLDH (n=300) within the same follow-up period.
Multivariate analysis was performed using patient baseline clinical and radiography data. Non-
linear, multivariate, logistic regression analysis was used to build a predictive model.
RESULTS: Recurrence of LDH occurred within 1 to 48 months after microdiscectomy. Preopera-
tively, patients who developed rLDH were smokers (70% vs. 27%, p<.01; odds ratio [OR]=6.31, 95%
confidence interval [CI]: 3.27–12.16) and had higher body mass index (29.0±6.1 vs. 27.0±4.3, p=.03;
OR=1.09 per 0.01 unit change). Radiological parameters that were associated with rLDH were higher
disc height index (0.35±0.007 vs. 0.26±0.002, p<.001), higher segmental range of motion (9.8±0.28°
vs. 7.6±0.11°, p<.001; OR=0.53 per 0.01 unit change), and lower central angle of lumbar lordosis
(33.4±0.81° vs. 47.1±0.47°, p<.001; OR=0.53 per 0.01 unit change). Additionally, Pfirrmann grade

FDA device/drug status: Not applicable.
Author disclosures: EB: Grant: Russian Science Foundation (F, Paid to
the institution), pertaining to submitted manuscript. AVK: Grant: Russian
Science Foundation (F, Paid to the institution), pertaining to submitted manu-
script. ESB: Grant: Russian Science Foundation (F, Paid to the institution),
pertaining to submitted manuscript. MBG: Grant: Russian Science Foun-
dation (F, Paid to the institution), pertaining to submitted manuscript. MCP:
Grant: Russian Science Foundation (F, Paid to the institution), pertaining
to submitted manuscript; Other: Barrow Neurological Foundation (F, Paid
to the institution), Women’s Board of the Barrow Neurological Institute (E,
Paid to the institution), Newsome Family Endowment in Neurosurgery (E,
Paid to the institution), pertaining to submitted manuscript. VAB: Grant:
Russian Science Foundation (F, Paid to the institution), pertaining to sub-
mitted manuscript.
The disclosure key can be found on the Table of Contents and at
www.TheSpineJournalOnline.com.
The authors have no relevant financial interests.
The study was performed with financial support from the Russian Science
Foundation, Grant No 15-15-30037.
* Corresponding author. Barrow Neurological Institute, St. Joseph’s
Hospital and Medical Center, c/o Neuroscience Publications, 350 W. Thomas
Rd, Phoenix, AZ 85013, USA. Tel.: (602) 406-3593; fax: (602) 406-4104.
E-mail address: Neuropub@dignityhealth.org (V.A. Byvaltsev)
1
These authors contributed equally to this work.

http://dx.doi.org/10.1016/j.spinee.2016.10.011
1529-9430/© 2016 Elsevier Inc. All rights reserved.
 E. Belykh et al. / The Spine Journal 17 (2017) 390–400 391

3 (OR=16.62, 95% CI: 8.10–34.11), protrusion type of LDH (OR=5.90, 95% CI: 3.06–11.36), and
Grogan sclerosis grades 3 and 4 (OR=4.81, 95% CI: 2.50–9.22) were also associated with rLDH.
Multivariate non-linear modeling allowed for more accurate prediction of rLDH (90% correct prediction
of rLDH; 99% correct prediction of no rLDH) than other univariate logit models.
CONCLUSIONS: Preoperative radiographic parameters in patients with LDH can be used to assess
the risk of recurrence after microdiscectomy. The multifactorial non-linear model provided more ac-
curate rLDH probability estimation than the univariate analyses. The software developed from this
model may be implemented during patient counseling or decision making when choosing the type
of primary surgery for LDH. © 2016 Elsevier Inc. All rights reserved.

Keywords: Complications; Intervertebral disc degeneration; Intervertebral disc displacement; Prevention; Radiography;
Statistics; Surgery

Introduction sible association with recurrence include developmental

Removal of lumbar disc herniation (LDH) is one of the
most frequent elective surgical procedures performed for de-
generative lumbar spine lesions. However, because of the
constant presence of unsatisfying outcomes of open
discectomies, neither surgeons nor patients are truly satis-
fied. Studies with long postoperative follow-up have shown
that almost 20% of patients develop postdiscectomy com-
plications requiring at least one additional surgery [1–3]. The
10-year results of the Maine Lumbar Spine Study showed that
25% of patients required reoperation within a median time
of 24 months after discectomy [4]. The spectrum of possi-
ble unfavorable outcomes includes the formation of epidural
fibrosis presenting with recurring radicular pain, postopera-
tive instability of the spinal motion segment, reherniation in
the early postoperative period, and recurrence of LDH [5–7].
Reherniation is defined as a repeated LDH occurring within
the first 6 months after discectomy, whereas recurrence is
defined as occurring more than 6 months after discectomy
[8]. Recurrent disc herniation and reherniations are the most
common reasons for secondary surgery during the first 2 years
after discectomy, and the rate of reoperations is 15% to 25%
[9,10]. Moreover, revision surgery is usually more compli-
cated than the initial operation due to postoperative adhesions.
The percentage of successful outcomes compared with primary
surgery also decreases [11]. The 10-year results of the Maine
Lumbar Spine Study showed that only half (26 out of 51, 51%)
of the patients who underwent subsequent surgery were sat-
isfied with their current state [4]. Additionally, after subsequent
surgeries, patients were less satisfied [12] and had worse dis-
ability and work outcomes [4]. Moreover, after the first
reoperation, patients had a considerable 25% cumulative risk
for further spinal surgery [3]. Reherniation and LDH recur-
rence are the most frequent conditions causing pain recurrence
and requiring subsequent surgical reoperation.
Since the mid-1990s, research has primarily been focused
on the detection of correlations between individual factors
and recurrence of lumbar disc herniation (rLDH) after primary
surgical treatment. Investigation has shown that intraopera-
tive findings and the degree of annular competence after
discectomy have a predictive value for reherniation and
reoperation [13]. Other factors that have been studied for pos-
abnormalities, extensive physical exercise, weight, smoking
[14], gender, age less than 35 years [15], stage of interver-
tebral disc degeneration, segmental instability, herniation type,
subligamentous disc herniation [15], intervertebral disc height,
diabetes [14], disc protrusion [14], and others [16–18]. None-
theless, some studies resulted in opposing conclusions
concerning these risk factors. There is a relative paucity of
studies that assess preoperative radiological predictors of rLDH,
and no work, to our knowledge, that presents a predictive
system that includes multiple radiological characteristics.
Knowledge of the preoperative predictors of recurrence would
be of significant value in clinical decision making. In this study,
we assessed the preoperative magnetic resonance imaging
(MRI) and radiographic parameters and investigated their in-
fluence on LDH recurrence after discectomy. Furthermore,
we created a calculator for estimating the probability of LDH
recurrence.
Materials and methods
Study design
The institutional ethics committee approved the study.
We conducted a retrospective case-control study and ob-
tained information about the patients who underwent
discectomy from January 2008 to July 2012. Patients who
developed rLDH that required reoperation within 3 years of
follow-up after the primary surgery composed Group 1. We
then randomly selected 300 patients (Group 2, control) who
underwent discectomy during the same period and did not
develop rLDH within 3 years of follow-up. The control
cohort was matched by the level of herniation (150 patients
with L4–L5 and 150 patients with L5–S1 LDHs) and pre-
operative diagnosis.
The inclusion criteria for both groups were the follow-
ing: LDH at the L4–L5 or L5–S1 level, and single-level
primary discectomy. The exclusion criteria for both groups
were the following: a combination of LDH with degenera-
tive stenosis of the spinal canal, degenerative or isthmic
spondylolisthesis, and traumatic lesions of the lumbar spine.
Patients with recurrence of pain due to stenosis of the spinal
canal, segmental instability without reherniation, peridural fi-
392 E. Belykh et al. / The Spine Journal 17 (2017) 390–400
Context
Recurrent disc herniation following treatment for lumbar
radiculopathy is a known issue. At present, risk factors for
the development of this condition are not well under-
stood. The authors evaluated this using data from 350
patients treated at their center.
Contribution
Fifty patients were found to have a recurrent disc hernia-
tion. A multivariable model used by the authors identified
a number of significant predictors including smoking status,
Pfirrmann grade and BMI among others.
Implications
The findings should be interpreted in light of a low number
of recurrences and a statistical model that is likely overfit
as a result. By standard convention, co-variates in a sta-
tistical model can be supported by every 20 patients with
the outcome of interest. At best then, given the low number
of recurrences in the authors’ cohort, only 2-3 variables
could really be supported in a multivariable model. The
number of co-variates included may lead to spurious results
unique to the patients and significant only by chance. This
directly impairs the translational capacity of these find-
ings. The study design cannot account for patients lost to
the system who may have sought care elsewhere for a re-
currence, nor can selection bias or preferential nonoperative
care be addressed. The evidence presented should be viewed
as Level IV as a result.
—The Editors
brosis, LDH at a different level, or contralateral rLDH were
excluded from the analysis.
The preoperative MRI and radiographic parameters of pa-
tients were assessed using data from the radiological archive
by two independent physicians who were blinded to the
outcome (recurrent LDH). Any disagreement in the grading
was discussed until a consensus was achieved.
MRI variables
The stage of intervertebral disc degeneration was as-
sessed on preoperative T2-weighted MRIs (1.5T MRI, Hitachi
Ltd, Tokyo, Japan) using the Pfirrmann grading system
(Table 1) [19]. Lumbar disc herniations were classified into
three types—protrusions, extrusions, or sequestrations—as de-
scribed by Fardon and Milette [20]. Modic changes of vertebral
end plates and bone marrow of adjacent vertebrae were also
assessed on T1-weighted and T2-weighted MRI scans [21].
Degeneration of the facet joint cartilage and the degree of sub-
chondral sclerosis of the facet joints were assessed as described
by Grogan et al. [22].
Table 1
Classification of Pfirrmann grade from magnetic resonance imaging [19]
Grade 1 Bright white, clearly defined nucleus pulposus, normal height
Grade 2 Bright with bands, clearly defined nucleus pulposus, normal
height
Grade 3 Gray, nucleus pulposus is not clearly defined, normal or
minimally decreased height
Grade 4 Gray/black, no nucleus pulposus, normal to moderately
decreased height
Grade 5 Black, no nucleus pulposus visible, severely decreased height
Radiographic variables
Lumbar radiography (Definium 8000, General Electric
Medical Systems, Waukesha, WI, USA) was performed in the
neutral anteroposterior and neutral lateral planes with the
patient standing. Additionally, functional lumbar radio-
graphs were obtained at the maximal forward bending position
and maximal extension position. Disc height index (DHI), sag-
ittal segmental range of motion (sROM), and the central angle
of lumbar lordosis were also assessed on plain lateral radio-
graphs. Disc height index was defined as the quotient of the
disc height to the height of the overlying vertebral body. Sag-
ittal segmental range of motion was defined as the difference
between the segmental angles on lateral flexion and exten-
sion radiographs. The segmental angle was measured on plain
lateral radiographs as the angle between the adjacent end plates
of the diseased spinal level. The central angle of lumbar lor-
dosis was measured on plain lateral radiographs using the Cobb
method [23] as the angle between the lines parallel to the su-
perior end plates of the L1 and S1 vertebrae. The presence
of retrolisthesis and lumbarization or sacralization of the lower
vertebrae was also recorded and assessed.
Statistical analysis
The statistical analysis was performed using SPSS 15.0
(IBM Corp, Armonk, NY, USA) and Statistica 8.0 (StatSoft
Inc, Tulsa, OK, USA) software. Quantitative data are pre-
sented as means and standard deviations. The non-
parametric Mann-Whitney U test and chi-square test were
used to perform comparisons between the groups. The
threshold level of statistical significance was set at p<.05.
Correlations among the variables were assessed using the
Spearman coefficient (r). Multiple logistic regression anal-
ysis with the binomial (yes or no) dependent variable of
rLDH within 3 years was conducted using Statistica 13.0
(Dell Software, Aliso Viejo, CA, USA). All variables deter-
mined to be significant by the Mann-Whitney U test and
chi-square test were used as variables in the model. The
Wald statistic and p value for each variable were then used
as tests of significance of the regression coefficients in the
model. All insignificant variables were removed. This was
done to ensure the model was not overparameterized. Finally,
we developed a simple calculator in Microsoft Excel
(Redmond, WA, USA) that estimates LDH recurrence prob-
ability based on the defined preoperative variables.
 E. Belykh et al. / The Spine Journal 17 (2017) 390–400 393

the 50 patients who developed rLDH composed Group 1. Re-

Fig. 1. Timing of recurrent disc herniation in Group 1 patients (n=50). (The
figure is used with permission from Barrow Neurological Institute, Phoenix,
Arizona.)
Results
Patient characteristics
There were 1,368 discectomies performed at the L4–L5
and L5–S1 levels for patients with nerve root compression
caused by LDH from January 2008 through October 2012.
All patients had planned surgery for LDH with a history of
at least 6 weeks of ineffective conservative treatment. Op-
erations were performed at the L4–L5 level in 57.7% (n=790)
and the L5–S1 level in 42.3% (n=578) of cases. Lumbar disc
herniation removal was performed according to the Caspar
open microdiscectomy technique described previously [24].
All operations were performed by neurosurgeons specializ-
ing in spinal surgery with significant experience in the
technique. We identified 78 patients who required revision
surgery within the follow-up period of up to 3 years. Of these,
currence of lumbar disc herniation occurred from 1 to 48
(12.6±12.0) months after the primary surgery. Another 28 pa-
tients were excluded from the analysis: 6 patients developed
contralateral rLDH, 7 patients had postoperative instability
and peridural fibrosis, and 15 patients developed postoper-
ative stenosis at the level of surgery. The distribution of
reherniation according to the time after discectomy showed
several peaks, the first during the first 3 months, after 6 to
12 months, and then at about 20 months after the initial surgery
(Fig. 1). Comparative analysis of the subgroup of patients who
developed rLDH within 6 months after surgery and the sub-
group of patients who developed rLDH more than 6 months
after surgery showed no significant differences in all studied
variables (Supplementary Appendix S2). Three hundred pa-
tients from the pool of 1,290 who had 3 years of follow-up
without rLDH and who did not have exclusion criteria com-
posed the control group (Group 2).
The female-to-male ratio and mean age were compara-
ble in the groups of patients with and without rLDH. Patients
with rLDH had a significantly higher body mass index (BMI)
(p=.026). When the data on preoperative symptoms were com-
pared, we found significantly lower preoperative back (p<.001)
and leg pain (p=.002) in the group of patients with rLDH,
although this difference may not be clinically significant
(Table 2). Patients with rLDH had a mean preoperative
Oswestry disability score of 51.8±17 comparable with the
mean preoperative score of those without rLDH, 48.3±15.8,
p=.09. The proportion of smokers was more than 2.5 times
greater in the group of patients who developed rLDH (70%,
35 out of 50) compared with that in the non-recurrent group
(27%, 81 out of 300) (p<.01), and the correlation with the
recurrence was r=.32 (p<.05). The average BMI was signifi-
cantly higher in the rLDH group (29.0±6.1 vs. 27.0±4.3,
p=.03), but the correlation with recurrence was weak, r=.12,
p<.05. Of the patients’ preoperative parameters, smoking status

Table 2
Preoperative patient characteristics according to recurrence of lumbar disc herniation
Group 1 Group 2
All patients Recurrent disc herniation Without recurrent disc herniation p Value r Value†
No. of patients 350 50 300
Age (years) 42.6±11.8 (17–77) 43.6±7.9 (30–61) 42.5±12.3 (17–77) .34 .05
Gender ratio (M:F) 161:189 (46%:54%) 22:28 (44%:56%) 139:161 (46%:54%) .76 −.02
BMI (kg/m2) 27.3±4.7 (17.9–45.8) 29.0±6.1 (18.3–45.8) 27.0±4.3 (17.9–42.5) .03 .12*
Preoperative VAS score in back 3.8±1.2 (1–7) 3.1±1.1 (1–5) 3.9±1.2 (1–7) <.01 −.19*
Preoperative VAS score in leg 6.0±1.5 (3–9) 5.4±1.4 (3–8) 6.1±1.5 (3–9) <.01 −.16*
Patients with preoperative neurologic deficit 286 (82%) 40 (80%) 246 (82%) .73 −.02
Duration of pain before surgery (mo) 4.2±4.6 (1–60) 3.5±6.1 (1–11) 4.4±4.9 (1–60) .66 −.02
Patients with right-sided radicular symptoms 169 (48%) 20 (40%) 149 (49.6%) .20 −.07
Preoperative Oswestry score (points) 48.8±15.9 (12–96) 51.8±17 (12–96) 48.3±15.8 (24–88) .09 .09
Smoking (yes/no) 116 (33%) 35 (70%) 81 (27%) <.01 .34*
BMI, body mass index; SD, standard deviation; VAS, visual analog scale.
The data are presented as means±SD (range), n (%), or as absolute value (%) for binomial variables. p Values are calculated using the Mann-Whitney U
test and χ2 criteria for binomial variables.
* p<.05 for the corresponding r.
†
r=Spearman correlation coefficient of a variable with lumbar disc herniation recurrence.
394 E. Belykh et al. / The Spine Journal 17 (2017) 390–400

Table 3
Correlation of radiological parameters with rLDH by group
Group 1 Group 2
Recurrent disc herniation Without recurrent disc herniation p-Value r-Value†
Level of LDH .60 .02
L4–L5 23 150 (50%)
L5–S1 27 150 (50%)
DHI, mm (mean±SD) 0.35±0.007 0.26±0.002 <.001 .53*
sROM, degrees (mean±SD) 9.8±0.28 7.6±0.11 <.001 .36*
Central angle of lumbar lordosis, degrees (mean±SD) 33.4±0.81 47.1±0.47 <.001 −.53*
Modic changes .10 .09‡
Type 1 31 (62.0%) 37 (12.3%)
Type 2 5 (10.0%) 89 (29.7%)
None 14 (28.0%) 174 (58.0%)
Pfirrmann disc degeneration <.001 .49*
Grade 3 38 (76.0%) 48 (16.0%)
Grade 4 12 (24.0%) 252 (84.0%)
Herniation type <.001 .32*
Protrusion 35 (70.0%) 85 (28.3%)
Extrusion 13 (26.0%) 210 (70.0%)
Sequestration 2 (4.0%) 5 (1.7%)
Grogan subchondral sclerosis <.001 .27*
Grade II 15 (30.0%) 202 (67.3%)
Grade III 33 (66.0%) 92 (30.7%)
Grade IV 2 (4.0%) 6 (2.0%)
Grogan degeneration of joint cartilage <.001 .29*
Grade II 7 (14.0%) 169 (56.3%)
Grade III 33 (66.0%) 103 (34.3%)
Grade IV 10 (20.0%) 28 (9.4%)
Lumbarization 5 (10.0%) 3 (1.0%) <.001 .21*
Retrolisthesis 19 (38.0%) 34 (11.3%) <.001 .26*
DHI, disc height index; rLDH, recurrence of lumbar disc herniation; SD, standard deviation; sROM, sagittal segmental range of motion.
* p<.05 for the corresponding r.
†
r=Spearman correlation coefficient of a variable with LDH recurrence.
‡
Spearman correlation is not significant; however, if the Modic change equals “none,” it is excluded from the analysis, Spearman r=−.48 and p<.05.

and BMI were included in the multivariate logistic regres-
sion analysis.
Radiological parameters
The results of comparisons of preoperative MRI and ra-
diographic parameters (LDH level, DHI, sROM, lordosis angle,
Modic changes, Pfirrmann grade, herniation type, Grogan
grades, lumbarization, and retrospondylolisthesis) of the lumbar
spine are shown in Table 3. Correlation analysis showed that
nine variables, namely DHI, sROM, the central angle of lumbar
lordosis, Pfirrmann grade, the type of herniation, Grogan scle-
rosis grade, Grogan degeneration grade, lumbarization, and
retrolisthesis, have significant correlations with the rLDH.
Grogan sclerosis and degeneration grades had significantly
high correlation (r=.76, p<.05) between each other.
However, we decided to include this variable in the model
because it increased the correct prediction rate of the whole
model for rLDH (Fig. 2). Among the univariate models, the
model with DHI as a single effect had the highest correct pre-
diction rate: 93.4% (327 out of 350) overall, 60% (30 out of
50) for the cases with rLDH, and 99.0% (297 out of 300) for
the cases without rLDH (Fig. 3). The sagittal lumbar lordo-
sis angle had a significant correlation with the DHI and was
also a significant predictor of rLDH (Fig. 4).
The resultant multiple non-linear logistic regression anal-
ysis yielded a mathematical model for rLDH chance estimation
with overall 98% correct prediction rate (Table 5). The model
had the following form:
p=
(
exp β0 + ∑ j =1 β j x j
N
)
( )
exp β0 + ∑ j =1 β j x j +1
N

Multiple non-linear logistic regression model

For the multiple logistic regression analysis, we selected
seven variables: BMI, DHI, sROM, the central angle of lumbar
lordosis, smoking, LDH type, and Pfirrmann grade. The re-
gression analysis showed that each variable was a significant
predictor of rLDH with the exception of sROM (Table 4).
where p is the probability, β0 is the intercept coefficient
and β is the variable coefficient, and x is the variable
magnitude. Finally, the classification-of-case analysis showed
that the model was able to predict 45 out of 50 (90%) of re-
currences and 298 out of 300 (99.3%) of cases without rLDH
correctly. The calculator for patient-specific rLDH probabil-
 E. Belykh et al. / The Spine Journal 17 (2017) 390–400 395

Table 4
Estimation of parameters in multiple non-linear logistic regression analysis with modeled probability that recurrence equals either 1 (event occurred) or 0
(event did not occur) (distribution of dependent variable is binomial, linked function is logit)
Effect (x) Level of effect β Standard error Lower 95% CI Upper 95% CI Wald statistic p Value
Intercept – 17.03 (β0) 8.85 −0.32 34.38 3.70 .054
BMI – −0.47 0.16 −0.79 −0.15 8.42 <.01
DHI – −115.75 36.57 −187.43 −44.07 10.02 <.01
sROM – 0.70 0.40 −0.08 1.49 3.09 .079
Central angle of lumbar – 0.61 0.18 0.27 0.96 12.16 <.01
lordosis
Smoking No 2.39 0.84 0.75 4.03 8.16 <.01
LDH type Extrusion 3.57 1.30 1.01 6.12 7.50 <.01
LDH type Protrusion −0.90 0.88 −2.61 0.82 1.05 .306
Pfirrmann Grade 3 −1.82 0.72 −3.22 −0.41 6.42 <.01
BMI, body mass index; DHI, disc height index; sROM, sagittal segmental range of motion; LDH, lumbar disc herniation; CI, confidence interval.

ity was generated using Excel and can be found in
Supplementary Appendix S1.
Discussion
Large population-based studies have demonstrated the risk
of reoperation for LDH to be 10% to 15% [3,4]. A Finnish
study of 35,309 discectomy patients has shown that 14% of
patients underwent secondary spinal surgery [3]. After the sec-
ondary surgery, patients had a 25% cumulative risk of further
spinal surgery during the next 10 years [3]. Although our study
was not designed to study the overall incidence of rLDH, we
identified 50 out of 1,368 patients (3%) who developed LDH
recurrence at the L4–L5 and L5–S1 levels within the 3-year
follow-up period. This rate is lower than the reported 15% to
20% of rLDH, possibly because we analyzed only cases of

Fig. 2. Graph of logistic regression analysis showing the sagittal, segmental range of motion as the independent variable, and recurrence as the binomial
(1—yes, 0—no), dependent variable. (The figure is used with permission from Barrow Neurological Institute, Phoenix, Arizona.)

Fig. 3. Graph of logistic regression analysis with the disc height index as the independent variable, and lumbar disc recurrence as the binomial (1—yes, 0—no),
dependent variable. (The figure is used with permission from Barrow Neurological Institute, Phoenix, Arizona.)
396 E. Belykh et al. / The Spine Journal 17 (2017) 390–400

Fig. 4. Graph of logistic regression analysis with the central angle of lumbar lordosis as the independent variable, and lumbar disc recurrence as the bino-
mial (1—yes, 0—no), dependent variable. (The figure is used with permission from Barrow Neurological Institute, Phoenix, Arizona.)

rLDH, but not all reoperations, and because some patients
with rLDH could have been missed on the long-term follow-up.
When assessing correlations of the variables with rLDH
in the subgroup of patients with early reherniations, we found
Spearman r values similar to the whole group of patients with
rLDH, with the exception of BMI (Supplementary
Appendix S2). Only when all patients were combined to-
gether did BMI have a significant, but low-powered, correlation
with recurrence. The BMI correlation with recurrence was
insignificant in subgroups when assessed separately. Our results
suggest that subgroups of patients with early and late
reherniations after open microdiscectomy may have similar
risk factors.
Non-modified patient risk factors
Swartz and Trost reported that age, gender, smoking status,
the level of herniation, and duration of symptoms were not
associated with rLDH [7]. Based on the observations of 44
pairs of homozygotic twins with a weight difference of more
than 8 kg, Videman et al. found that the cumulative load of
larger body weight slightly delayed the process of disc de-
generation [25]. Fritsch et al. conducted a retrospective analysis
of 182 medical records of patients who underwent second-
ary operations and concluded that age was not a reliable
prognostic risk factor of rLDH; however, only 17 out of 98
(18%) male patients had a satisfactory long-term outcome com-
pared with 12 out of 38 (32%) female patients, p<.05 [26].
Dai et al. analyzed data from 39 patients who underwent
reoperation for rLDH and concluded that factors such as age,
gender, traumatic events, number of prior surgeries, level of
herniation, side of recurrence, pain-free interval, duration of
recurrent symptoms, walking capacity, and preoperative Jap-
anese Orthopaedic Association score had no prognostic
significance for rLDH [27]. Our study also did not reveal dif-
ferences in patients’ age, gender, and side of herniation
between the groups of patients with and without rLDH. Al-
though differences in preoperative visual analog scale scores
for leg and back pain were significant (p<.01 for both), the
absolute clinical value of such differences is questionable.
Smoking and BMI
Smoking has been proven to impair postoperative wound
repair in many surgical studies [28]. It is believed that toxic
substances cause vasoconstriction and decrease nutrient supply

Table 5
Comparison of different models that estimate chances of recurrent lumbar disc herniation (rLDH) using Statistica software
Prediction based on the model
Model Odds ratio Lower 95% CI Higher 95% CI % Correct rLDH % Correct non-rLDH
Multiple logistic regression N/A N/A N/A 90.00 99.33
BMI (per 0.01 increase)‡ 1.09* N/A N/A 0 100.00
DHI (per 0.01 increase)‡ N/A* N/A N/A 60.00 99.00
sROM (per 1 degree increase)‡ 1.89* N/A N/A 20.00 99.33
Central angle of lumbar lordosis (per 1 degree increase)‡ 0.67* N/A N/A 56.00 96.33
Smoking (positive status) 6.31† 3.27 12.16 0 100.00
LDH type (protrusion) 5.90† 3.06 11.36 4.00 98.33
Pfirrmann (grade 3) 16.62† 8.10 34.11 76.00 84.00
BMI, body mass index; DHI, disc height index; sROM, sagittal segmental range of motion; LDH, lumbar disc herniation; CI, confidence interval; N/A,
not available.
* Odds ratio calculated using univariate logistic regression analysis.
†
Odds ratio calculated using free odds ratio calculator; % correct calculated using discriminant analysis (https://www.medcalc.org/calc/odds_ratio.php).
‡
Logistic regression model.
 E. Belykh et al. / The Spine Journal 17 (2017) 390–400
to the intervertebral disc, as well as diminish the activity of
cellular tissue repair mechanisms [29]. Our study further sup-
ports the existing data [30] that smoking is associated with a
higher risk of rLDH. Nonetheless, we doubt that persistent
smoking status, as a single predictor, would influence clin-
ical decision making regarding a change of treatment tactics.
However, we believe that in multivariable risk assessment,
smoking status adds validity to the rLDH odds estimation.
Obesity (BMI>30) was also shown to be an independent
risk factor for rLDH after microdiscectomy [31]. Our study
also showed a weak but significant correlation (r=.12, p<.05)
between BMI and rLDH. This correlation is stronger (r=.23,
p<.5) in the cohort of patients with BMI >30 than in the cohort
with BMI ≤30.
Cessation of smoking is frequently recommended before
surgery. However, the time required for the risks to de-
crease to a level equivalent to the non-smoking population
after smoking cessation is still unknown. The BMI, al-
though alterable in the long term, is also difficult to change
before surgery.
Intraoperative variables
The risk for rLDH was shown to be significantly higher
in fragmented herniations with an annular defect (9 out of
33, 27%), compared with the fragment or fissure (1 out of
89, 1%), fragment contained (4 out of 42, 10%), or no frag-
ment contained (2 out of 16, 12%) types of LDH [13].
Differences in intraoperative details of LDH removal in each
type of LDH result in the differences in the competence of
the annulus. Such decreased annular competence leads to sig-
nificantly higher reherniation and reoperation rates for patients
with larger annular defects. We have previously demon-
strated that different microdiscectomy techniques lead to
variations in complication profiles and different risks for rLDH
[8]. In the current study, LDHs were classified as protru-
sion, extrusion, or fragmented herniations. Interestingly, our
study showed that the protrusion type of LDH is found more
often in the cohort of patients who develop rLDH (70%, 35
out of 50) than in the control group (28%, 84 out of 300).
The estimation of the size of the annular defect in different
herniation types is beyond the scope of the current study.
Radiological variables
Kim et al. have examined the radiological risk factors for
rLDH after microdiscectomy at the L4–L5 levels in the great-
est detail [32]. The study revealed no significant correlation
between the disc degeneration grade and recurrence. However,
the authors demonstrated that the incidence of rLDH was lower
in discs with Pfirrmann degeneration grades I and II (1 out
of 33, 3%) and VI (0 out of 13, 0%) than in discs with grades
III, IV, and V (13 out of 111, 12%). According to this study,
factors that have significant correlation with rLDH include
smoking, increased DHI, and increased segmental sROM [32].
Our study further reinforces these findings. We identified that
397
rLDH significantly correlates with the following param-
eters: DHI, Pfirrmann disc degeneration grade, type of
herniation, sROM, central angle of lumbar lordosis, and type
I Modic changes of adjacent vertebral bodies.
The question of whether all patients would benefit from
preoperative flexion-extension lumbar spine radiography
remains unanswered. Although it may be unnecessary in some
cases, we found this diagnostic method useful for our prac-
tice, as it may provide additional information about the
dynamic stability of the spinal segments.
Disc degeneration grade
The association of Pfirrmann grade with the risk of rLDH
was previously suggested by Kim et al. [32] and confirmed
in our study. Only Pfirrmann grades 3 and 4 were recorded
in our study, and grade 3 had higher odds of recurrence than
grade 4. Pfirrmann grade 4 is associated with disc collapse,
and such patients either had not presented with LDH or were
treated with fusion and were not included in our study.
Pfirrmann grade 1 is roughly normal and is not characteris-
tic of LDH, whereas grade 2 discs have clear differentiation
between the nucleus and annulus with normal or slightly de-
creased disc height. Our findings are in line with the
biomechanical model of LDH, where herniations were re-
producibly created by cyclic loading in mildly degenerated
disc specimens but not in the moderately and severely de-
generated discs [33].
Disc height and segmental range of motion
The results of this study are aligned with the results of basic
biomechanical studies of LDH development [34,35]. In bio-
mechanical ex vivo models, LDH was produced by a highly
compressive load at high flexion angles [34,35] due to the
rupture of collagen fibers of the annulus at the site of attach-
ment to the vertebral end plate [36]. End plate changes may
also contribute to decreased collagen fiber attachment to the
annulus and may explain the increased risk of rLDH. The study
findings indicate that increased DHI and increased sROM are
significant risk factors of rLDH. From the biomechanical point
of view, larger disc height before discectomy could lead to
a higher disc height decrease after nucleus removal and in-
creased segmental mobility. DHI and sROM were significantly
correlated (r=.78, p<.05) in our study, confirming the bio-
mechanical link between these parameters. A decrease in disc
height could be observed as a naturally occurring process of
age and progression of degeneration. Restoration of disc height
is critical for foraminal decompression and overall sagittal
spinal alignment.
Lumbar lordosis
The angle of lumbar lordosis had an insignificant corre-
lation with sROM and low negative correlation with DHI
(r=−.17, p<.05). Lower lumbar lordosis, ie, flat lumbar spine,
is associated with biomechanically suboptimal sagittal spinal
alignment and distribution of body weight. A decrease in
lumbar lordosis significantly increases the risk of disc
398 E. Belykh et al. / The Spine Journal 17 (2017) 390–400
degeneration and primary LDH [37]. Our study also showed
a significant reduction of lumbar lordosis in the cohort of pa-
tients with rLDH. This decrease in lumbar lordosis was
associated with the protrusion type of LDH (r=−.25, p<.05)
when all 350 patients were analyzed, although the power of
the correlation was weak and became insignificant in the sub-
group analysis.
Facet joint degeneration
Facet joint cartilage degeneration grade and cartilage scle-
rosis grade correlated well with each other (r=.76, p<.05) but
had a low-power correlation with rLDH (r=.29 and r=.27, re-
spectively, p<.05). Both grades were significantly correlated
with sROM (r=.31 and r=.30, respectively, p<.05), confirm-
ing the association of degenerative spinal instability with the
changes in the facet joints [38].
Multivariate logistic regression model
We performed a multivariate logistic regression analysis
to assess the odds of rLDH. The regression analysis allowed
us to identify a set of predictive factors that may play an in-
fluential role in the development of rLDH. Our study
elaborated on a valuable clinical hypothesis that a risk for
rLDH could be predicted from patient characteristics. The gen-
eralized non-linear regression model was developed to aid in
quantitative odds estimation. Knowledge of the objective risk
rate plays an essential role in deciding on the method and
extent of surgical treatment for LDH during patient consul-
tation. If a high risk of rLDH is identified, the surgeon can
plan ahead for preventive methods and modify the surgical
procedure accordingly. For example, several approaches could
be considered to prevent rLDH in high-risk patients, such as
lumbar fusion [39], annular repair [40], or implantable devices,
to close the annular defect [41].
The algorithm and rLDH risk calculator that was devel-
oped allows for the preoperative assessment of the rLDH risk.
We have implemented this scoring system in our practice and
believe that this system improves clinical decision making
for patients with LDH. However, this model needs to be tested,
preferably in a multicenter, prospective, adequately powered
trial. The question of whether the use of this preoperative in-
formation could change how clinical care is delivered and
whether it could decrease the incidence of rLDH remains
unanswered.
Study limitations
This study has several limitations. Our multiple logistic
regression analysis did not include some relevant risk factors
that were identified as significant independent predictors in
other studies. For example, we did not examine the size of
the annular defect. The association of large annular defects
with rLDH was previously investigated elsewhere [13]. The
annular defect variable is typically assessed intraopera-
tively, and annulus repair strategies are implemented to
decrease the risk of reherniation [40]. An inherent draw-
back of the case-control study design is the uncertainty about
the causation of the risk factors. Thus, we built our rLDH risk
calculator based on the assumption of the causality of the
factors shown to be significantly different between the groups.
Additionally, we assessed data from only 350 patients of
the entire cohort of 1,368 patients. All 50 patients with rLDH
out of 78 who underwent reoperation were included in Group
1, whereas only 300 (out of 1,290) initially treated patients
were included in Group 2. Such factors as availability of com-
plete patient follow-up information and the necessity to
manually assess the images archived drove us to choose this
study methodology. Such selection could lead to biases if the
selected patients are different from the general cohort of pa-
tients without rLDH. We assume that random patient selection
decreased the chances for bias and that the 300 patients without
rLDH composed a representative control group. Also, this study
did not assess the difference in complication rates between
the surgeons.
We did not assess patients who underwent discectomy at
the L3–L4 level because there were no reherniations at this
level. Thus, our results should be extrapolated to the L3–L4
level with caution.
Studies that combine intraoperative factors such as annular
defect size and the amount of disc material removed have been
suggested, but the predictive values of such intraoperative pa-
rameters can be determined only during surgery. There are
several drawbacks to intraoperative predictors. First, if in-
traoperative parameters indicated a high risk of recurrence,
the surgeon would not have much time to adjust or change
the surgical plan. Second, the intraoperative predictors could
not be known at the time of the preoperative patient counseling.
Quantitative longitudinal parameters such as DHI, sROM,
and the central angle of lumbar lordosis could have
interobserver differences between the studies. Further studies
should also be done to validate the interobserver and
intraobserver accuracy of such measurements.
Conclusions
Seven parameters were significant predictors of rLDH at
the level of the previous microdiscectomy: BMI, DHI,
hypermobility of the spinal motion segment, flattening of
lumbar lordosis, smoking, disc protrusion, and Pfirrmann grade
III disc degeneration. The non-linear logistic regression model
we developed for the estimation of the chances of rLDH based
on the multivariate assessment of risk factors was signifi-
cantly more satisfactory than any single predictor model and
was able to predict 90% of rLDH correctly in the study. This
study further reinforces the evidence that preoperative ra-
diological parameters can be used to predict the risk of rLDH
after microdiscectomy. Surgeons could use this information
to advise patients of their risk for reherniation and as a premise
for selective use of various approaches aimed to prevent re-
currence, such as annulus suturing and barrier implants. Further
studies are needed to evaluate the model with other sets of
patients.
 E. Belykh et al. / The Spine Journal 17 (2017) 390–400
Acknowledgments
The authors are thankful to the Neuroscience Publica-
tions office at Barrow Neurological Institute and to Liudmila
Shcherbakova, MD, for help with preparation of the manu-
script. The authors are grateful for funding from the Barrow
Neurological Foundation, the Women’s Board of the Barrow
Neurological Institute, and, in part, the Newsome Family En-
dowment in Neurosurgery to Dr Preul. The authors declare
no relevant competing financial interests. The funding agen-
cies had no role in the study design, data collection and
analysis, decision to publish, or preparation of the manuscript.
Supplementary material
Supplementary material related to this article can be found
at http://dx.doi.org/10.1016/j.spinee.2016.10.011.
References
[1] Atlas SJ, Keller RB, Chang Y, Deyo RA, Singer DE. Surgical and
nonsurgical management of sciatica secondary to a lumbar disc
herniation: five-year outcomes from the Maine Lumbar Spine Study.
Spine 2001;26:1179–87.
[2] Malter AD, McNeney B, Loeser JD, Deyo RA. 5-year reoperation rates
after different types of lumbar spine surgery. Spine 1998;23:814–
20.
[3] Osterman H, Sund R, Seitsalo S, Keskimaki I. Risk of multiple
reoperations after lumbar discectomy: a population-based study. Spine
2003;28:621–7.
[4] Atlas SJ, Keller RB, Wu YA, Deyo RA, Singer DE. Long-term
outcomes of surgical and nonsurgical management of sciatica secondary
to a lumbar disc herniation: 10 year results from the Maine Lumbar
Spine Study. Spine 2005;30:927–35.
[5] Axelsson P, Karlsson BS. Intervertebral mobility in the progressive
degenerative process. A radiostereometric analysis. Eur Spine J
2004;13:567–72.
[6] Kraemer R, Wild A, Haak H, Herdmann J, Krauspe R, Kraemer J.
Classification and management of early complications in open lumbar
microdiscectomy. Eur Spine J 2003;12:239–46.
[7] Swartz KR, Trost GR. Recurrent lumbar disc herniation. Neurosurg
Focus 2003;15:E10.
[8] Belykh E, Giers MB, Preul MC, Theodore N, Byvaltsev V. Prospective
comparison of microsurgical, tubular-based endoscopic, and
endoscopically assisted diskectomies: clinical effectiveness and
complications in railway workers. World Neurosurg 2016;90:273–
80.
[9] Mobbs RJ, Newcombe RL, Chandran KN. Lumbar discectomy and the
diabetic patient: incidence and outcome. J Clin Neurosci 2001;8:10–13.
[10] Ozgen S, Naderi S, Ozek MM, Pamir MN. Findings and outcome of
revision lumbar disc surgery. J Spinal Disord 1999;12:287–92.
[11] Bridwell KH, Anderson PA, Boden SD, Vaccaro AR, Wang JC. What’s
new in spine surgery. J Bone Joint Surg Am 2011;93:1560–6.
[12] Fritzell P, Knutsson B, Sanden B, Stromqvist B, Hagg O. Recurrent
versus primary lumbar disc herniation surgery: patient-reported
outcomes in the Swedish Spine Register Swespine. Clin Orthop Relat
Res 2015;473:1978–84.
[13] Carragee EJ, Han MY, Suen PW, Kim D. Clinical outcomes after
lumbar discectomy for sciatica: the effects of fragment type and anular
competence. J Bone Joint Surg Am 2003;85-A:102–8.
[14] Huang W, Han Z, Liu J, Yu L, Yu X. Risk factors for recurrent lumbar
disc herniation: a systematic review and meta-analysis. Medicine
(Baltimore) 2016;95:e2378.
399
[15] Yurac R, Zamorano JJ, Lira F, Valiente D, Ballesteros V, Urzua A. Risk
factors for the need of surgical treatment of a first recurrent lumbar
disc herniation. Eur Spine J 2016;25:1403–8.
[16] Cinotti G, Roysam GS, Eisenstein SM, Postacchini F. Ipsilateral
recurrent lumbar disc herniation. A prospective, controlled study. J Bone
Joint Surg Br 1998;80:825–32.
[17] Kim JM, Lee SH, Ahn Y, Yoon DH, Lee CD, Lim ST. Recurrence
after successful percutaneous endoscopic lumbar discectomy. Minim
Invasive Neurosurg 2007;50:82–5.
[18] Suk KS, Lee HM, Moon SH, Kim NH. Recurrent lumbar disc
herniation: results of operative management. Spine 2001;26:672–6.
[19] Pfirrmann CW, Metzdorf A, Zanetti M, Hodler J, Boos N. Magnetic
resonance classification of lumbar intervertebral disc degeneration. Spine
2001;26:1873–8.
[20] Fardon DF, Milette PC. Nomenclature and classification of lumbar disc
pathology. Recommendations of the combined task forces of the North
American Spine Society, American Society of Spine Radiology, and
American Society of Neuroradiology. Spine 2001;26:E93–113.
[21] Modic MT, Steinberg PM, Ross JS, Masaryk TJ, Carter JR.
Degenerative disk disease: assessment of changes in vertebral body
marrow with MR imaging. Radiology 1988;166(1 pt 1):193–9.
[22] Grogan J, Nowicki BH, Schmidt TA, Haughton VM. Lumbar facet joint
tropism does not accelerate degeneration of the facet joints. AJNR Am
J Neuroradiol 1997;18:1325–9.
[23] Cobb JR. Outline for the study of scoliosis. In: Thomson JEM, Blount
WP, editors. Instructional course lectures, The American Academy of
Orthopaedic Surgeons. Ann Arbor, MI: JW Edwards; 1948. p. 261–75.
[24] Caspar W. A new surgical procedure for lumbar disc herniation causing
less tissue damage through a microsurgical approach. In: Wüllenweber
R, Brock M, Hamer J, Klinger M, Spoerri O, editors. Lumbar disc adult
hydrocephalus. Berlin: Springer-Verlag; 1977. p. 74–80.
[25] Videman T, Gibbons LE, Kaprio J, Battie MC. Challenging the
cumulative injury model: positive effects of greater body mass on disc
degeneration. Spine J 2010;10:26–31.
[26] Fritsch EW, Heisel J, Rupp S. The failed back surgery syndrome:
reasons, intraoperative findings, and long-term results: a report of 182
operative treatments. Spine 1996;21:626–33.
[27] Dai LY, Zhou Q, Yao WF, Shen L. Recurrent lumbar disc herniation
after discectomy: outcome of repeat discectomy. Surg Neurol
2005;64:226–31, discussion 31.
[28] Rinker B. The evils of nicotine: an evidence-based guide to smoking
and plastic surgery. Ann Plast Surg 2013;70:599–605.
[29] Porter SE, Hanley EN Jr. The musculoskeletal effects of smoking. J
Am Acad Orthop Surg 2001;9:9–17.
[30] McGirt MJ, Ambrossi GL, Datoo G, et al. Recurrent disc herniation
and long-term back pain after primary lumbar discectomy: review of
outcomes reported for limited versus aggressive disc removal.
Neurosurgery 2009;64:338–44, discussion 44–5.
[31] Meredith DS, Huang RC, Nguyen J, Lyman S. Obesity increases the
risk of recurrent herniated nucleus pulposus after lumbar
microdiscectomy. Spine J 2010;10:575–80.
[32] Kim KT, Park SW, Kim YB. Disc height and segmental motion as risk
factors for recurrent lumbar disc herniation. Spine 2009;34:2674–8.
[33] Wilke HJ, Ressel L, Heuer F, Graf N, Rath S. Can prevention of a
reherniation be investigated? Establishment of a herniation model and
experiments with an anular closure device. Spine 2013;38:E587–93.
[34] Adams MA, Hutton WC. Prolapsed intervertebral disc. A hyperflexion
injury 1981 Volvo Award in Basic Science. Spine 1982;7:184–91.
[35] Veres SP, Robertson PA, Broom ND. The morphology of acute disc
herniation: a clinically relevant model defining the role of flexion. Spine
2009;34:2288–96.
[36] Wade KR, Robertson PA, Thambyah A, Broom ND. “Surprise” loading
in flexion increases the risk of disc herniation due to annulus-endplate
junction failure: a mechanical and microstructural investigation. Spine
2015;40:891–901.
[37] Ergun T, Lakadamyali H, Sahin MS. The relation between sagittal
morphology of the lumbosacral spine and the degree of lumbar
400 E. Belykh et al. / The Spine Journal 17 (2017) 390–400
intervertebral disc degeneration. Acta Orthop Traumatol Turc
2010;44:293–9.
[38] Jaumard NV, Welch WC, Winkelstein BA. Spinal facet joint
biomechanics and mechanotransduction in normal, injury and
degenerative conditions. J Biomech Eng 2011;133:071010.
[39] Wang JC, Dailey AT, Mummaneni PV, et al. Guideline update for the
performance of fusion procedures for degenerative disease of the lumbar
spine. Part 8: lumbar fusion for disc herniation and radiculopathy. J
Neurosurg Spine 2014;21:48–53.
[40] Bailey A, Araghi A, Blumenthal S, Huffmon GV. Prospective,
multicenter, randomized, controlled study of anular repair in lumbar
discectomy: two-year follow-up. Spine 2013;38:1161–9.
[41] Parker SL, Grahovac G, Vukas D, et al. Effect of an annular closure
device (Barricaid) on same level recurrent disc herniation and disc height
loss after primary lumbar discectomy: two-year results of a multi-center
prospective cohort study. Clin Spine Surg 2016; doi:10.1097/
BSD.0b013e3182956ec5 [e-pub ahead of print].