Correlation of expertise with error detection skills of force application during spinal manipulation learning*

Motor learning is of crucial importance in many health disciplines such as dentistry, chiropractic, and physical therapy.^1,2  Clinical training in manual therapies using high-velocity low-amplitude spinal manipulation (HVLA-SM) procedures, similar to other motor skills, is modulated by intrinsic (e.g., central nervous system maturation, physical capabilities) and extrinsic factors (practice schedule, knowledge of performance, and knowledge of results).³  The rate of HVLA-SM skill development using various combinations of laboratory and instructional procedures has been measured, and several studies have identified parameters such as preload force, which is the gradual force applied to the spine before the thrust, time-to-peak force (impulse duration), peak force, and thrust rate as predictors of performance and expertise.^3–8  Most studies investigating HVLA-SM learning have used either human subjects or instrumented devices (manikins, force plates, strength gauges) and clearly demonstrated the relevance of including motor learning strategies as soon as possible in the chiropractic curriculum.^3,8–12 

It has been suggested that practice is a key element of motor skill acquisition and that the amount of practice can increase the strength of the recognition memory when paired with appropriate feedback (knowledge of results or knowledge of performance).¹³  Recent studies have confirmed that extrinsic feedback from a teacher and visual or auditory information provided with an instrumented teaching device improves HVLA motor skills in novices.^3,11  Major outcomes of practice include the ability to produce movement in a more efficient manner¹⁴  and the improved capability of learners to evaluate their own motor performance.¹⁵ 

One's ability to evaluate his or her own performance can be evaluated through error detection capability. For instance, experts' performance is characterized by improved monitoring of sensory feedback that allows them to use performance-related information to perceive an error. Detecting an error requires an accurate perception of the speed, force, distance, or duration of the movement and the ability to compare the action to a predetermined reference.¹⁶  Learning a specific motor task allows the development of a reference of correctness (memory trace) that is compared to the executed movement, allowing detection and correction of the movement in subsequent attempts.^17,18  In a study by Schmidt and White,¹⁹  subjects were asked to move using a ballistic-timing task, a slide with a follow-through over a predetermined distance, with a target movement time of 150 milliseconds. Subjects were then asked to estimate their movement duration after each trial. To quantify the error detection capabilities of subjects, the estimates were subtracted from the real movement time (subjective error) and correlated to the real movement time subtracted from the target movement time (objective error). The hypothesis was that an increased correlation between the subjective and objective errors would reflect a better error detection mechanism. Results from the study indicated that the correlation increased from 0.28 to 0.76 over a session of 120 practice trials. One of the known benefits of practice is that learners, particularly when they achieve high levels of proficiency, will become less dependent on external feedback (instructors and instrumented feedback), concomitantly relying on improved error detection capability to assess their own performance.

To our knowledge, no previous study has been conducted on error detection capabilities of force application while performing HVLA-SM or HVLA simulations. Therefore, the goals of this study were to evaluate if expertise leads to a more efficient movement and if error detection skills in HVLA simulation force application improve with expertise. Based on previous studies,^3,8  it was hypothesized that biomechanical parameters of spinal manipulation would improve with the level of expertise and that the error detection skills are significantly better in experienced clinicians.

Sixty-three participants involved in a chiropractic program favoring a mixed pedagogical strategy with positioning training and complete practice of HVLA-SM skills voluntarily participated in the current study. Given that there were no data available to conduct a standard sample size calculation, the sample size estimate was based on similar studies investigating expertise in spinal manipulation.^8,20,21  All participants gave their informed written consent according to the protocol approved by the Ethics Committee Research Involving Humans at Université du Québec à Trois-Rivières (CER-14-201-07.17). Participants were all tested during the same week during a brief (15–20 minutes) session conducted in the institution's research laboratory.

Four different groups of participants were tested based on their clinical experience and expertise. Group 1 (N = 15; mean age = 20.3) exclusively consisted of 1st-year students of the chiropractic program who had no previous experience with HVLA-SM or HVLA simulation and only theoretical notions related to positioning and HVLA-SM techniques. Group 2 (N = 17; mean age = 23) comprised 4th-year students, with 3 years of supervised clinical practice. Group 3 (N = 15; mean age = 24.2) included 5th-year students with 4 years of supervised clinical practice. Participants in group 4 (N = 16; mean age = 36.8) were all experienced chiropractors with clinical postgraduation experience ranging from 2 to 44 years of clinical practice. Participants' characteristics are presented in Table 1.

Participants were instructed to complete 10 consecutive trials of their best thoracic spine manipulations on an instrumented device using a right-handed pisiform contact. The maneuver, identified as a prone unilateral hypothenar transverse push adjustment, was performed with a posterior-to-anterior force vector (relative to the instrument) using either a left or right contact and the body positioning of their choice. All participants were instructed to perform the HVLA-simulated procedure using preload, and a demonstration trial was performed by the same author (JT) to properly illustrate and identify the procedure and the contact point. No feedback regarding their performance was provided for these first 10 trials. Participants were then asked to perform 15 practice trials of the same HVLA simulation technique with the specific goal of reaching a maximum peak force of 300 N. This value is inferior to those reported in the literature and was selected to ensure that 1st-year students could reach it.⁴  This familiarization phase was conducted so that all subjects could gauge the level of resistance offered by the device and learn how to reach the target force. After each trial, participants were shown their force-time curve, and the maximum force reached was provided both verbally and visually. Finally, participants were asked to complete 10 additional trials and still try to reach 300 N without being provided with feedback. After each trial, participants were asked to give an estimate of the maximal peak force that was reached. All task requirements and all instructions provided to the participants were identical for all 4 groups of participants.

Spinal manipulations were performed on a computer-connected device developed to simulate a thoracic spine and to record obtained force-time profiles for HVLA simulation analysis. Participants placed their hands on a device contact point linked to a strain gauge (model UL 400, Statham Inc, Oxnard, CA) by a spring that allowed replicating movements and resistance of the thoracic spine. Applied vertical forces over the contact point were recorded with the strain gauge and displayed in real time on the computer screen. In order to simulate the articular release characterizing a vertebral joint cavitation, a moving piece (the contact point, the spring, and the strain gauge) was set to drop down 5 mm when a threshold force parameterized from the computer was reached during HVLA simulations (set to 300 N). The target force was set at 300 N so that all participants could easily reach this value regardless of their expertise or any mechanical and anthropometric disadvantages. Before performing spinal manipulation, an electromagnet (model EM300-24-212, APW Company, Rockaway, NJ) fixed to the device was activated by the computer to attract the lower component of the device and move it up in position. During the manipulation, a device circuit compared the strain gauge–measured force with the threshold force every millisecond. When the measured force exceeded the threshold force, the electromagnet was switched off and the lower portion of the device dropped down. An acquisition card (NI USB-6008, National Instruments, Austin, TX) connected by a USB cable between the device and the computer allowed data collection using Labview software (National Instruments, Austin, TX).

The force-time signals were analyzed to determine the onset of force, peak force applied, and preload force. All parameters were analyzed for each trial and each participant. The data were then used to calculate time-to-peak force (thrust duration) and rate of force application. Three measures of mean error for peak force were calculated: subjective error (SE), objective error (OE), and variable error (VE). For each experimental trial, peak force was measured as the highest force value reached in the force-time curve. Using this information, SE, OE, and VE in peak force were calculated and compared for each trial and each participant. SE measures the participants' ability to accurately estimate their peak force and represents the positive or negative difference between the actual and estimated peak forces. A positive SE corresponds to an underestimation of the force reached, and a negative SE corresponds to an overestimation of the force reached. The participants' ability to accurately reach the target force of 300 N was termed OE and represents the target force subtracted from the peak force reached. VE measures the inconsistency in peak force and represents the difference between the participants' peak force score on each trial and their own average score. Thus, if participants generated peak force very consistently, their VE tended to be small.

Peak force data (SE, OE, and VE) and the 4 basic biomechanical parameters were analyzed with 1-way analysis of variance (ANOVA) separately for each variable. Effect size was calculated using partial eta-squared values and the significance level was set at p < .05 for all analyses. The statistical analysis was performed using Statistica 10 (StatSoft, Inc, Tulsa, OK).

ANOVA assumptions (normal distribution of data and homogeneity of variances) were met, and the analyses yielded significant between-group differences for each of the 4 basic biomechanical parameters of HVLA simulations. A significant difference in preload forces (F [3, 561] = 35.18, p < .0001, ηp² = .15) was observed between 1st-year students and the 3 other groups (4th and 5th year, experts). Time-to-peak force values significantly decreased (F [3, 561] = 28.14, p < .0001, ηp² = .13) from 1st-year students to experts, with significant differences observed between each group except the 4th- and 5th-year students. The thrust force significantly increased (F [3, 561] = 17.33, p < .0001, ηp² = .09) from the 1st- to the 5th-year students, with values similar to the experts observed in the 4th and 5th year. Finally, a rate of force application significantly increased with experience (F [3, 561] = 38.7, p < .0001, ηp² = .17). Figure 1 illustrates the mean values for preload, time-to-peak force, peak force, and thrust rate for all 4 groups.

With regard to error-detecting capabilities, no significant between-group differences were found for any of the 3 variables of interest (for all, p > .05). Mean (SD) values for SE, OE, and VE are presented in Table 2.

The objectives of this study were twofold: firstly, to evaluate if biomechanical parameters of HVLA simulation improve with the level of expertise, and secondly, to assess if error detection capability during HVLA simulation force application improves with expertise. Overall, the results indicate that there is a significant improvement in all 4 basic biomechanical parameters from the 1st-year student group to the expert group. The large increase observed in preload force from the 1st- to the 4th-year participants could be explained by the fact that the 1st-year students had no formal practice of HVLA-SM and that this concept of preload force was unfamiliar to them. As shown in a previous study, preload forces are quite variable during the first 2 years of training, and definite patterns of improvement are difficult to identify.⁸  Time-to-peak force decreased steadily with the level of expertise, whereas the peak force increased for all 3 groups of students, typically characterizing improvements in HVLA simulation skills throughout practice. Similar observations have been reported in previous studies confirming that HVLA simulation biomechanical parameters improve with clinical experience and that the participants involved in the present study represent typical chiropractic students and expert chiropractors.^3,8,12 

The present study failed to identify any between-group difference in the error detection capability variables. The lack of significant expertise effect in error detection capability could be related to the traditional teaching methods used in the training of HVLA-SM procedures. For instance, learning to detect and correct one's own error is an important skill in itself that requires practice. To the best of our knowledge, error detection is not an intricate component of HVLA-SM training, and students, although they learn to produce various force levels, are rarely provided with objective quantified feedback related to force application. Thus, one could argue that all the participants in this study were novices in the interpretation of sensory feedback related to errors. Consequently, regardless of their clinical expertise, experts and trainees were not familiar with such information and how this information could be used to generate appropriate corrections in the following HVLA simulation attempts. A study by Robertson et al.²²  comparing novice and skilled gymnasts walking across a balance beam as quickly as possible with and without vison of the beam identified clear differences between the 2 groups. Indeed, their results showed that experts, adequately correcting movement errors, were less affected by the elimination of vision than novice beam-walkers. One possible explanation for the lack of expertise effect in HVLA error detection is that, as opposed to novice and expert chiropractors, novice and expert gymnasts are trained to detect their movement errors and to implement corrections as they progress through years of training.

Given the usual organization of practice and learning sessions in HVLA-SM training, error detection capability may not currently be an adequate variable to assess expertise since error-detecting skills are not formally included in training. Providing augmented feedback to trainees has been shown to improve performance³  and could be used to implement a learning procedure that emphasizes error detection and related movement corrections.

In order to assess the quality and the generalizability of the acquired processes, future research should investigate complementary indicators of expertise such as the transfer of learning of 1 specific task to variations of this task. Generalization or transfer of motor learning refers to individuals applying what has been learned in one context to another that shares similar characteristics.^23,24  Such a variable may be better suited in chiropractic training, since daily practice of HVLA-SM involves high levels of adaptation and contextualization of the task (table height, patient's morphology, clinical status).

This study demonstrates that clinical experience and level of training seem to be important factors leading to improvement in HVLA-SM and HVLA simulation skills. As for error detection capabilities, no significant difference was found between students and experts. Feedback based on error detection could be implemented in chiropractic programs to allow future clinicians to increase their skills at detecting errors and perhaps improve procedure safety.

This study was funded through the Chaire de recherche en chiropratique-FRCQ and the Institut Franco-Européen de Chiropraxie (IFEC). The funding sources had no role in study design, data collection, analysis, data interpretation, or writing of the manuscript.