Introduction

Transpedicle screw fixation has been accepted worldwide since Harrington et al. first placed pedicle screws through the isthmus [1]. Studies have reported that transpedicle screw fixation has gained an advantage over other forms of spinal instrumentation [2, 3]. Safety concerns on the violation of the spinal canal leading to potential harm to vascular, neural, and other vital structures have been encouraging surgeons to improve the accuracy of pedicle screw placement by trying various approaches [2, 3, 760]. Through many conventional methods, usually relying on anatomical landmarks, pedicle screws were inserted with or without fluoroscopy to localise the pedicle. However, studies have reported high rates of cortical perforation [13, 24, 27, 53].

Image-guided technology has greatly broadened the scope of modern surgery and has improved the clinical results of various non-spinal orthopaedic operations [46]. In vivo and in vitro studies indicated that pedicle screw insertion accuracy could be significantly improved with image-assisted systems compared with conventional approaches [10, 15, 21, 22, 24, 25, 27, 32, 36, 44]. Among them, computed tomography-based navigation (CT Nav) was the most popular. CT Nav could provide precise anatomy of the pedicle as well as reduced radiation exposure. However, new concerns about the system arose with a steep learning curve and excessive preoperative preparation including computed tomography with a specific protocol, data acquisition and transfer, and patient registration [23, 45]. The development of intraoperative two dimensional and three dimensional fluoroscopy-based navigation (2D FluoroNav and 3D FluoroNav) appeared to tackle such issues [7, 23, 25, 34, 38, 39, 45, 52]. The equipment did not require registration, it reduced imaging time and radiation dosage, and avoided repeated C-arm movements during surgery because visualisation of the surgical instruments in relation to the patient’s anatomy in all the desired image planes was determined from the beginning. Though a few studies implied that accuracy of fluoroscopy-assisted pedicle screw insertion was comparable with that of CT Nav [23, 43, 45], different population characteristics and assessment methods of placement accuracy in various studies resulted in inconsistent conclusions.

To investigate these issues, the authors conducted a systematic meta-analysis of the published literature which looked at the accuracy of pedicle screw placement with the image-guided systems in the human spine. Postoperative methods used for pedicle screw placement assessment were identified, and we used a uniform method to assess the insertion accuracy in this study. A meta-analysis combining the results of numerous studies could obtain more value for clinical reference than a single study.

Materials and methods

Literature search

A literature search was conducted covering the span from 1990 until March 7, 2009 using OVID, Springer, and MEDLINE databases. Screening included titles, subtitles, and abstracts by combining the term pedicle screw with each of the following keywords: computer assisted/assistance/aided, image guided/guiding/guidance, navigation, and navigated. No language restriction was used. Additionally, a review of all references from the identified papers was performed.

The following inclusion criteria were used: (1) the study should have a design in which the pedicle screw was inserted with one of the navigation methods (studies without an image-assisted study design were excluded), (2) the study should include living or cadaveric human spine (studies with only animal or artificial models were excluded), and (3) postoperative assessment of the pedicle screw placement was conducted.

Meta-database construction

To construct the meta-database, the fields and parameters as defined below, when available, were recorded for each study. Categories were developed to classify studies as either clinical and/or cadaveric. The demographics of each study population were recorded when available and applicable. Navigation types and methods for postoperative assessment of the pedicle screw placement of each study were discriminated. The range of vertebral segments instrumented with pedicle screws was recorded for single spinal level (e.g. the cervical, thoracic, lumbar [including S1]), multi-spinal levels (e.g. cervical and thoracic, thoracic and lumbar, the whole spine) and spinal level unstated for each study. The total number of pedicle screws and misplaced screws were noted. Data of studies including multi-spinal levels were extracted and recorded according to detailed spinal level if they were available, and were used for further subgroup analysis. Figure 1 depicts how the studies were classified. In this study, a unified way was used for assessing pedicle screw accuracy. Pedicles without any perforation after screw insertion were extracted from the total screws placed and were defined as nonbiased screws. Studies comparing two or more navigation techniques or different populations (e.g. in vivo patient vs cadaveric, scoliosis vs non-scoliosis) were divided into sub-studies.

Fig. 1
figure 1

Schematic showing the subgroups developed from the meta-database based on subject type and spinal level

After setting up the meta-database, subgroups were analysed according to the filtering of results. The meta-database was first divided based on the subject type (in vivo patient or cadaveric population) and then separated based on the spinal levels. If multiple examiners and/or multiple postoperative assessing methods were applied in the paper, the most accurate way was retained (e.g. results stored radiologist over those of the surgeon, and direct dissection results over CT and CT over radiographic assessments).

Meta-analysis

The descriptive statistics in this study summarise the data using SPSS 10.0 software (SPSS Inc., Chicago, IL, USA). The descriptive statistics included 5% trimmed mean, the median, standard deviation, minimum and maximum, and interquartile range. The 5% trimmed mean was calculated by excluding the lowest and highest 2.5% of the data (possible outliers). The median is used to calculate the central tendency for both normal and skewed distributed data. The box plots display the 25th and 75th percentiles (interquartile range) of the data set, the median accuracy, and possible outliers.

Results

Literature search

The search resulted in 131 references from the database. Seventy-seven studies (58.8%) were omitted in accordance with our predefined exclusion criteria. Thirty-five papers were removed because they were unrelated to image-guided pedicle screw insertion. Eighteen of the excluded studies were the review papers. Four surveys were conducted on either animal or artificial models. Two technical note papers and four studies merely addressing radiation dose or registration issues were also omitted. Two experiments on robot-assisted navigation were not taken into the final review. Twelve clinical articles were removed because pedicle screws were placed but no information could be used for accuracy assessment. Thus, data of 54 [760] remaining studies were extracted for further analysis.

Meta-database characteristics

The remaining 54 studies (48 English, four German, one French, and one Chinese) could be classified into 35 clinical [741] and 20 cadaveric [4160] articles. One study [41] was labelled into two categories. In the clinical group, all studies used postoperative CT for their analysis of pedicle screw placement. Two of them used CT and radiographs, and one study applied CT and EMG along with intraoperative visualisation. In the cadaveric group, six researches (30.0%) used dissection-based approaches, four (20.0%) used CT, seven (35.0%) used CT and dissection, and three studies used CT and radiographs (15.0%) for assessment. In the in vivo studies, the most common assessment method for pedicle placing accuracy was the 2-mm increment deviation classification. Twenty-five studies (71.4%) used this classification along with the screw direction when deviation was present. However, in the in vitro tests, various assessments were identified based on the authors’ own definitions.

According to navigation type, all 54 studies could also be classified into 40 CT navigated, 20 2D fluoroscopy navigated, and six 3D fluoroscopy navigated sub-studies. Five studies (two with CT Nav and three with 2D FluoroNav) were divided because each of them contains two populations with remarkably different characteristics. Seven studies could be classified into two categories because they used two navigation methods in their experiments.

Based on spinal level, 30 studies (55.6%) included a single spinal level and 20 studies (37.0%) contained two spinal levels. The spinal levels of the remaining four studies (7.4%) were unstated. Of the 20 studies containing two spinal levels, twelve studies were further divided into sub-studies based on extracted single spinal level because they provided enough information about the pedicle screw number in the specific spinal level.

In our review, two cadaveric studies [43, 45] compared screw insertion accuracy between the CT and 2D fluoroscopy-guided methods. In three retrospective in vivo studies [7, 23, 38], two navigation methods were compared (2D vs 3D FluoroNav, 2D FluoroNav vs CT Nav, and 3D FluoroNav vs CT Nav, respectively).

A total number of 7,533 pedicle screws were summarised in our database with 6,721 screws accurately inserted into the pedicles (89.22% weighted mean accuracy). A total of 6,063 were placed in vivo, with a weighted mean accuracy of 90.21% (median 91.14%), and 1,470 were placed in vitro, with a weighted mean accuracy of 85.10% (median 90.83%).

In vivo population filter and descriptive statistics

In 42 in vivo studies or sub-studies, there were 27 CT navigated, ten 2D fluoroscopy navigated, and five 3D fluoroscopy navigated articles. For the CT Nav subgroup, 3,554 screws were placed and the range in placement accuracy was 26.22% (minimum 72.03%, maximum 98.25%) with a median accuracy of 90.76%. In the 2D FluoroNav subgroup, a total of 1,219 pedicle screws were inserted. The median accuracy and range were 85.48% and 23.32% (minimum 72.73%, maximum 96.05%), respectively. For the 3D-fluoroscopy navigated subgroup, the range in placement accuracy was 18.44% (minimum 80.85%, maximum 99.29%) with a median accuracy of 97.16%. Further descriptive statistics and a box plot comparing these three subgroups can be found in Table 1 and Fig. 2.

Table 1 Descriptive statistics of the pedicle screw insertion accuracy in the in vivo and cadaveric populations at all spinal levels with the assistance of different navigation systems
Fig. 2
figure 2

Box plots showing the accuracy of pedicle screw insertion assisted by computed tomography (CT), 2D fluoroscopy, and 3D fluoroscopy-based navigation for all spinal levels of the in vivo patient

Based on pure studies (studies with two spinal levels were not divided), the CT navigated subgroup only at the lumbar and/or thoracic spine levels included 16 papers and a total of 2,333 pedicle screws placed. Using the same filtering for the 2D fluoroscopy navigated subgroup resulted in eight papers and a total of 1,156 pedicle screws placed with a lower median accuracy (86.51%) than the CT-assisted subgroup (91.14%). Descriptive statistics and a box plot for these subgroups can be found in Table 2 and Fig. 3, respectively.

Table 2 Descriptive statistics comparing the in vivo and cadaveric subgroups only at the lumbar and/or thoracic spine levels with the assistance of CT and 2D fluoroscopy-based navigation
Fig. 3
figure 3

Box plots comparing pedicle screw insertion accuracy with computed tomography (CT) and 2D fluoroscopy-based navigation for only the thoracic and/or lumbar spinal levels of the in vivo patient

Based on pure and spinal level extracted studies and filtering by spine level resulted in three pure thoracic, six extracted thoracic, six pure lumbar and four extracted lumbar papers for the in vivo subgroup with the use of CT navigation. Using the same filtering for the 2D fluoroscopy navigated group resulted in two pure thoracic, five extracted thoracic, one pure lumbar and four extracted lumbar papers. In the lumbar level, CT Nav obtained a median accuracy of 90.32% compared with 86.96% for the 2D FluoroNav subgroup. However, the discrepancy of median accuracy between the two navigational methods increased at the thoracic level (91.52% vs 82.51%). Further descriptive statistics can be found in Table 3.

Table 3 Descriptive statistics comparing only the in vivo subgroups at the lumbar or thoracic spinal level with the assistance of CT and 2D fluoroscopy-based navigation

Cadaveric population filter and descriptive statistics

In 24 in vitro studies or sub-studies, there were 13 CT navigated, ten 2D fluoroscopy navigated, and one 3D fluoroscopy navigated articles. For the CT Nav subgroup, 801 screws were placed and the range in placement accuracy was 50.00% (minimum 50.00%, maximum 100.00%) with a median accuracy of 94.59%. In the 2D fluoroscopy navigated subgroup, a total of 663 pedicle screws were inserted. The median accuracy and range were 90.12% and 50.04% (minimum 43.33%, maximum 98.37%), respectively. Further descriptive statistics and a box plot comparing these three subgroups can be found in Table 1 and Fig. 4.

Fig. 4
figure 4

Box plots comparing pedicle screw insertion accuracy with computed tomography (CT) and 2D fluoroscopy-based navigation for all spinal levels of the cadaveric population

Based on pure studies (studies with two spinal levels were not divided), there were ten papers and a total of 660 pedicle screws inserted in the CT navigated subgroup only with the lumbar and/or thoracic spine levels included. Using the same filter for the 2D fluoroscopy navigation group resulted in nine studies and a total of 633 pedicle screws placed with a lower median accuracy (91.67%) than the CT-based navigation group (94.65%). Descriptive statistics and a box plot for these subgroups can be found in Table 2 and Fig. 5, respectively.

Fig. 5
figure 5

Box plots comparing pedicle screw insertion accuracy with computed tomography (CT) and 2D fluoroscopy-based navigation for only the thoracic and/or lumbar spinal levels of the cadaveric population

Based on pure and extracted studies and filtering by spine level resulted in five pure thoracic, two extracted thoracic, one pure lumbar and two extracted lumbar papers for the in vivo subgroup with the use of CT navigation. Using the same filter for the 2D fluoroscopy navigated subgroup resulted in two pure thoracic, one extracted thoracic, three pure lumbar and one extracted lumbar papers. In the thoracic level, CT Nav obtained a median accuracy of 92.50% compared with 88.57% for the 2D FluoroNav subgroup. In the lumbar level, three studies with CT navigation obtained pedicles all accurately placed into the pedicle (100.00%) and the median accuracy was 94.17% for the 2D fluoroscopy navigation group. Further descriptive statistics can be found in Table 4.

Table 4 Descriptive statistics comparing only the in vitro subgroups at the lumbar or thoracic spinal level with the assistance of CT and 2D fluoroscopy-based navigation

Discussion

Although the transpedicle screw fixation is commonplace in spinal surgery, it is still a technically demanding procedure. Various conventional methods were developed to assist the screw insertion. However, the high pedicle malposition rate encouraged surgeons to search for the perfect method for screw placement [13, 24, 27, 53]. Image guided systems appeared to improve the surgical accuracy for pedicle screw placement. Kosmopoulos et al. [61] pooled previous studies on pedicle screw placement accuracy and used a descriptive method to assess the accuracy between conventional and navigated methods. The study indicated that navigation does provide a higher accuracy in the placement of pedicle screws except at the thoracic levels. In the studies of our search, a total of 13 in vivo [10, 12, 13, 15, 18, 21, 22, 24, 25, 27, 32, 35, 36] and nine in vitro [42, 44, 4851, 53, 55] studies or sub-studies were detected to compare the pedicle screw insertion accuracy between the conventional and navigational methods. We further analysed the relative risk of malposition rate of the image-guided over the conventional method and found most of them indicated that navigation groups were associated with a smaller risk of screw malposition compared with non-navigation groups with statistically significant RR values in one in vitro [44] and nine in vivo [10, 15, 21, 22, 24, 25, 27, 32, 36] studies. Overall, pedicle screw insertion with navigation assistance does improve accuracy especially in the in vivo studies.

The first generation of navigation applied to spinal surgery was the computer tomography-assisted system. Many researches indicated that CT Nav could improve pedicle screw placement accuracy [10, 15, 22, 24, 27, 32, 36, 44]. However, concerns about the system with a steep learning curve and excessive preoperative preparation prevented this technique from being widely adopted [23, 45]. The 2D and 3D FluoroNav systems appeared to solve the above problem by eliminating or prevented steps such as preoperative scanning, data acquisition and transfer, and intraoperative registration [23, 45]. In our searched studies, one in vivo retrospective study [23] compared pedicle screw insertion accuracy between 2D FluoroNav and CT Nav. CT Nav obtained a lesser but insignificant risk compared with the 2D FluoroNav (P = 0.45, RR = 0.58, 95% CI = 0.14–2.36). Two in vitro studies [43, 45] also compared the accuracy between 2D FluoroNav and CT Nav. The pooled estimate of effect size from the two studies demonstrated that the CT-based navigation group was associated with a lesser (but insignificantly so) risk of screw malposition rate compared with 2D FluoroNav group (P = 0.12, RR = 0.64, 95% CI = 0.36–1.12). Another in vivo retrospective study [38] compared accuracy between 3D fluoroscopy and CT-based navigation. 3D FluoroNav obtained a lesser (but insignificantly so) risk compared with the CT assistance (P = 0.15, RR = 0.71, 95% CI = 0.45–1.12). The statistically insignificant differences might be attributed to the small number of pedicle screws in these studies. As a result, it became necessary to perform a meta-analysis combining the results of numerous previous studies to draw more confident and acceptable conclusions, thus resulting in our study.

After reviewing the searched papers, we found only six studies [7, 25, 34, 38, 39, 45, 52] which assessed accuracy of the 3D fluoroscopy assisted pedicle screw insertion (five in vivo and one in vitro). Due to prevented studies with such assistance, the authors did not make much descriptive analysis for 3D FluoroNav. The main part of this survey focussed on 2D fluoroscopy and CT-based navigation and we tried to compare the screw insertion accuracy between them. Based on subject type, we found that the median placement accuracy for both the in vivo and in vitro subgroups with CT assisted navigation (in vivo = 90.76%, in vitro = 94.59%) was higher than that of the corresponding subgroup with 2D FluoroNav (in vivo = 85.48%, in vitro = 90.12%) when only pure studies with any spinal level were considered (Table 1, Figs. 2 and 4). It was also interesting to note that the results were found to be consistent when only thoracic and/or lumbar segments were considered (Table 2, Figs. 3 and 5). Moreover, when we further extracted single spinal level data from studies with multi-spinal levels and re-analysed the database based on one spinal level (thoracic or lumbar), the same conclusion was made, namely, that CT Nav provided a higher accuracy compared with 2D FluoroNav (Tables 3 and 4). It was noteworthy that discrepancy between the two navigation methods became larger after extracting single spinal level data from multi-spinal level studies, especially at the in vivo thoracic level (CT Nav = 91.52%, 2D FluoroNav = 82.51%) (Table 3), which meant there was less potential in the use of 2D FluoroNav than CT Nav in this spinal segment. However, the smaller difference between the two at the lumbar level implied a similar advantage in the use of both navigation methods.

Kosmopoulos et al. [61] discussed the importance of clearly defining the pedicle screw placement assessment method. Take the common assessment of “in” or “out” for example; a screw “in” the pedicle could be interpreted as a screw completely in the pedicle without any violation and also could be regarded as a screw safely placed in, allowing perforation to certain extent, which was mainly determined by the evaluator’s opinion. In our study, various kinds of assessment methods were identified from the reviewed papers and most studies provided detailed information about the magnitude and direction of misplacement. The author carefully reviewed the papers and extracted the information of screws perfectly in the pedicle without any deviation to build a comparable database allowing for synthesis of the results. Though simple, such gauging provided a unified method used to synthesise numerous published studies and draw a more acceptable result.

In this study we mainly used the median to describe the central tendency rather than the arithmetic mean, weight or geometric mean. For one thing, the database in our study does not represent results of one experiment or the conclusion of homogenous researches, and it’s better to “describe” rather than to “calculate”. As a result, we used the median along with box plots for our study. Another reason is that the median could be used to describe the central tendency for both normal and skewed distributed data, and it was possible to detect the difference between a group of normal data and another one of skewed data. The other information we provided in the study, such as 5% trimmed mean, was for the purpose of providing convenience for comparisons when other authors reported results with a similar form.

Theoretically, it is of convenience to control the variates in the cadaveric study which could provide confident results. However, such results were questioned on account of the difference between in vivo and in vitro conditions. As a result, RCTs comparing two medical treatments might provide more powerful evidence than other kinds of research for clinical studies. Though three retrospective studies [7, 23, 38] were identified, no RCTs were found comparing two navigation methods after our search, which might not lead to a powerful conclusion. As a result, the current meta-analysis might obtain clinical reference, for cumulative studies were taken into the analysis with a uniform assessing method.

There are limitations to this study. The best evidence should be a meta-analysis of numerous RCTs with high quality. Our research is a descriptive research on various studies with different population characteristics and methods. Heterogeneities might result from differences in surgeons’ skill, screw dimension/accuracy assessment method, and publication bias. Furthermore, it might not be proper to extract single spinal level data from multi-spinal level studies to make a data pool. However, we conducted an extensive search with no language limitation and a uniform method to obtain the data and pool an acceptable and comparable database. Detailed information provided by the studies made it possible to explore more potential information about the screws in the specific spinal level, and cumulative findings resulting from a meta-analysis with a specific spinal level can provide more value for clinical reference. It was noteworthy was that the difference in pedicle placing accuracy between 2D fluoroscopy and CT-assisted navigation remained consistent when compared with each other based on different classifications.