Complex regional pain syndrome (CRPS) is a complication after surgery or trauma, although spontaneous development has also been described. CRPS is characterized by signs and symptoms of inflammation and central sensitization. The diagnosis can be made using several different criteria sets, the most popular of which are the International Association of the Study of Pain (IASP) and the Bruehl criteria sets.1 There are two types of CPRS; type 1 without an obvious detectable nerve lesion (CRPS1) and type 2 with an obvious detectable nerve lesion (CRPS2).
The IASP criteria have a high sensitivity but a lower specificity, whereas the Bruehl criteria have a high specificity but a lower sensitivity. The IASP criteria are useful in the clinical setting, whereas the Bruehl criteria seem to be more useful for research purposes.2 New IASP criteria are under discussion3 and attempts have been made to obtain a less subjective diagnosis by using diagnostic tools such as three-phase bone scan, radiograph, magnetic resonance imaging, functional magnetic resonance imaging, and temperature measurement devices.4 Because of the limited validity of clinical diagnoses, it may be difficult to differentiate CRPS1 from other related diseases. It is assumed that a false-positive diagnosis for CRPS1 is over-expressed, especially in patients with unclear complaints of symptoms such as pain.2
Temperature is one of the variables used in the diagnosis of CRPS1. Surface temperature of an extremity reflects the result of a complex combination of centrally regulated and locally affected thermoregulatory systems. We previously described a calculation method to examine the difference between videothermographic pictures of CRPS1 patients and healthy controls.5 Only a few studies have reported on the diagnostic value of temperature differences at the early onset of CRPS1 using thermography. In 1998, Birklein et al.6 described the temperature development in CRPS1 after a fracture compared with healthy subjects using different sympathetic stress factors; Gradl et al.7 reported on 158 fracture patients of whom 18 developed CRPS; and Schurmann et al.8 investigated differences in sympathetic control of 50 CRPS1 patients compared with 50 normal fracture patients and 50 controls (no fracture); a high sensitivity and specificity was found. However, none of the above-described methods has been accepted as a “gold standard.” Thus, factors that need to be studied further are the sensitivity, specificity, positive predictive value, and negative predictive value. All this can be explored by comparing patients who develop CRPS1 after a fracture, the CRPS1 patients, to patients who develop signs and symptoms after a fracture similar to CRPS1 and the control patients with complaints. A third control group is added comprising patients after fracture who have no symptoms or signs of CRPS1, controls without complaints. Furthermore, to derive indices on diagnostic value, receiver operating characteristics (ROC) curves to calculate the diagnostic value should be used. Several other factors related to thermographic recordings and analyzing methods also warrant further study. On average, thermographic recordings consist of 2300 pixels each representing a temperature on one extremity, thus calculation methods that compare the whole temperature profile of both hands should be considered. In addition, none of the earlier studies compared the palmar/plantar side with the dorsal side. The present study focuses on the validity of static skin surface temperature recording, applying different mathematical methods, for diagnosing (acute) CRPS1 patients. The term “static thermography” refers to a thermographic recording of an extremity without application of any disturbing factors on temperature regulation of that extremity.
This study was approved by the Medical Ethics Committee of the Erasmus MC (MEC no 198.780/2001/24). All participants provided written informed consent.
Patients with various types of fractures were first seen in the emergency room at three hospitals, the Erasmus MC, the Medical Center Rijnmond-Zuid location south, and the Medical Center Rijnmond-Zuid location Clara (Fig. 1). All patients were treated with a plaster cast for 6 (IQR, 4–8) wk, depending on the type of fracture. A questionnaire on the symptoms of CRPS1 was completed by the plaster specialist on average 2 (IQR, 0–5) days after removal of the plaster. Excluded were patients younger than 18 yr and patients with demonstrable nerve damage in the fractured limb after CRPS2. In addition, patients unable to complete a Dutch-language questionnaire were also excluded. A questionnaire on the symptoms of CRPS1 was completed by means of a short interview that addressed the anamnesis part of symptoms of CRPS1 proposed by Bruehl et al.1 and by the IASP. The anamnesis was particularly focused on signs and symptoms of the criteria of Bruehl. (Questions were asked about pain, vasomotor, sudomotor, and motor/ trophic changes.) CRPS1 was considered to be present when patients had continuing pain, hyperesthesia, temperature asymmetry and/or skin color asymmetry, edema and/or sweating asymmetry, motor and/or trophic changes. When patients met four of five of the Bruehl criteria and/or four of four of the IASP criteria, they were referred to an anesthesiologist (F.J.P.M.H) who has wide experience with CRPS1 patients; this physician conducted a comprehensive physical examination, after which only the Bruehl criteria were noted for each patient. This resulted in three groups: 1) 24 fracture patients fulfilling the Bruehl criteria designated as the CRPS1 patients, 2) 84 fracture patients with various complaints but not fulfilling the Bruehl criteria designated as control patients with complaints, and 3) 12 randomly selected (normal healing) fracture patients without any visible signs/complaints designated as control patients without symptoms. To ensure that patients with pain but without CRPS1 did not develop CRPS1 after their first visit, a second visit was planned 8 wk later. After the first consultation with the physician, videothermographic images were recorded after a standard protocol. The physician was blinded for the thermographic recording, and the technician who performed the recordings was blinded for the physician’s diagnosis. Before the recording, patients were acclimatized in a room with a mean temperature of 23°C (range, 22.5°C–23.5°C) and a relative humidity of 50% (range, 45%–55%) for 15 min. Patients were placed in a chair in an upright position.
Measurements of the involved (fracture) and noninvolved (not fractured) extremity were performed on the palmar/plantar side and the dorsal side. The hands where placed in a plexiglas frame. The frame has positioning points between digit 1 and digit 2, and between digit 3 and digit 4, which allows recording of comparable parts of the extremity in different patients. Based on average temperature of the palmar/plantar side, >+0.3°C was considered warmer and <−0.3°C was considered colder.
The foot temperature was recorded using a support below the ankle, which enables recording of the plantar aspect of the foot; the dorsal aspect was recorded by placing the feet on the ground. To establish whether the temperature difference also spread outside of the fractured area, a thermographic recording depicting the front of the leg from the knee down to the ankle was recorded in the same upright position.
Skin temperature of both extremities was registered with a computer-assisted infrared thermograph (ThermaCAM SC2000, Flir Systems, Berchem, Belgium). This infrared thermographic camera has a resolution of 320 × 240 pixels. Each temperature value measured in the picture is represented by one pixel; this gives a total of 76,800 temperature values recorded in one image. The thermographic images were stored on a hard disk (ThermaCAM Researcher 2001 HS, Berchem, Belgium) for further analysis.
The distance between the camera and the hand being measured was adjusted to 68 cm; thereby the resolving capacity on the hand was 0.8 × 0.8 mm2. The distance between the camera and the feet was adjusted to 90 cm to accommodate the whole foot; thereby the resolving capacity on the feet was 1.2 × 1.2 mm2.
To obtain only those pixels that represent the hand or feet, the data are filtered by a threshold. On average, one hand is represented by 23,500 pixels and a foot by 12,000 pixels.
Most of the commonly used methods calculate differences in mean skin surface temperature between the involved (fractured) and the noninvolved (not fractured) extremity. However, these methods consider the total surface extremity, or only an arbitrary region of interest such as the fingertips/toes. We developed inhouse software using Matlab to facilitate these and newly developed calculation methods as described below.
Absolute Difference in Mean Hand or Foot Temperature
The difference between average hand/foot temperature was calculated for both the palmar/plantar side and dorsal side using the following formula:
Absolute Difference in Mean Fingertip Temperature
A square was placed around each finger and toe tip of the extremity. The zeros in the square, indicating background, were filtered out.
Absolute Static Temperature Difference Between Wrist/Ankle and Fingertip/Toe Tip
For the hands, 5 points at the wrist (base), 5 points at the knuckle of each finger, and 5 points at the tip of each finger were defined using software. On each hand or foot, 5 lines were automatically drawn by computer over the hand/foot, as shown in Figures 2a and b. Hereafter a line was fitted through the temperature point that lay on the five lines. The slope, calculated by the fit, of each line was used to calculate the temperature increase/decrease across each of the fingers. The increase/decrease of each finger on the involved site was subtracted from the increase/ decrease of the corresponding fingers on the noninvolved site. This result was summed to indicate a total difference in temperature increase/decrease between the extremities. The same procedure was applied for the calculation of foot temperatures. For this 5 points were defined at the ankle (base), 5 points at the base of each toe, and 5 points at the tip of each toe (Fig. 2b). The aim of this calculation is to give an indication of the difference in vasomotor tone between the involved and noninvolved side. A large decrease in temperature between wrist compared with finger tips or ankle compared with toe tip indicates a high vasomotor tone resulting in low blood flow through fingers and toes, whereas a small decrease in temperature indicates a low vasomotor tone.
This calculation method determines the asymmetry factor (correlation) between the temperature histogram of the involved and noninvolved extremity, based on the method described by Huygen et al.5 The asymmetry factor describes the degree of dissimilarity (expressed in correlation coefficient) between the temperature data obtained from the involved hand/foot compared with that from the noninvolved hand/foot. A score of 1 indicates a similar temperature distribution between involved and noninvolved side; a lower score indicates less similarity. This method intends to consider all aspects of the whole temperature profile in comparing hands.
This Euclidian distance is a measure of the distance between the temperature histograms of the involved and noninvolved side.9
D = Distance between histograms; Involved, Noninvolved = frequency value of a class; i = bin class number.
The class width was set to 0.1°C. This calculation effectively calculates the degree of similarity in the shape of the temperature histogram of the involved and noninvolved site. The above-described calculations were used to measure the similarity of the palmar/plantar side and the dorsal side between the involved and noninvolved extremity.
Total Temperature Difference Between Fingers and Toes
For the hands, 5 points at the wrist (base), 5 points at the knuckle of each finger, and 5 points at the tip of each finger were defined using software. On each hand or foot, five lines were automatically drawn by computer over the hand/foot, as shown in Figures 2a and b. The temperature profile of each line on the involved hand/foot was compared with noninvolved hand/foot using cross-covariance (mean-removed cross-correlation). The total difference was calculated by summing the maximum cross-correlation found on each finger, with a maximum shift of 20 pixels. This measurement intends to evaluate the irregularity in temperature that is found in most CRPS1 patients. This irregularity expresses itself in so-called “hot spots” and “cold spots”; this method is able to compare local temperature disturbance on that can be a result of, for example, local inflammation.
Data analyses were performed with SPSS 14.0. One-way analysis of variance (ANOVA) was used to calculate any significant differences in age and weeks after trauma between CRPS1 patients, control patients with complaints, and control patients without symptoms. Cross-tabs χ2 was used to test whether the signs showed a significant difference between CRPS1 patients and control patients with complaints.
One-way ANOVA (Bonferroni test correction) was used to test whether the outcomes of the calculation methods used on the thermographic data showed a significant difference between CRPS1 patients, control patients with complaints, and control patients without symptoms.
The ROC is used to calculated the diagnostic value. The ROC is a graph, which is a very good indicator of the discriminating power of a diagnostic method. The coordinates of the graph are defined by calculating the sensitivity and specificity at different values of the diagnostic test (in this article the various methods to calculate the temperature difference between extremities), so-called “cutoff points.” This results in a graph of the true positive rate (sensitivity) against the false positive rate (specificity) for the different possible cutoff points in a given diagnostic test. The most valid diagnostic cutoff value was chosen at the highest combination of sensitivity and specificity.
The area under the ROC is a measure of the accuracy of the diagnostic test at hand, expressed in area under the curve (AUC). The accuracy is measured by a five-point system: excellent (AUC of 1–0.9), good (AUC of 0.9–0.8), fair (AUC of 0.8–0.7), poor (AUC of 0.7–0.6), and fail (AUC of 0.6–0.5). More insight into the diagnostic value of thermography is gained when the positive and negative predictive values are also calculated. The positive predictive value is the proportion of patients with positive test results who are correctly diagnosed. The negative predictive value is the proportion of patients with negative test results who are correctly diagnosed. In all tests, a P value <0.05 was considered to be statistically significant.
Demographic data and characteristics of the study population are given in Table 1. No significant difference was found in the incidence of CRPS1 patients among the three participating hospitals. Control patients without complaints where significantly younger compared with the other two groups. There was no significant difference among the three groups in the number of weeks after trauma, or in the location of the fracture. In CRPS1 patients, the involved side was more often warmer than colder (18 vs 6) compared with the noninvolved side.
Data on symptoms according to the Bruehl criteria are given in Table 2. By definition, control patients without complaint had no symptoms nor signs. Although CRPS1 patients had a slightly higher pain score (5.9) than the control patients with complaints (4.8), the difference was not significant (P = 0.104). Because CRPS1 patients were included according to the Bruehl criteria, a 100% score in each category for each symptom was mandatory. The control patients with complaints also showed relatively high percentages on all symptoms, except for vasomotor signs. A marked increase was found comparing sensory and vasomotor signs of the CRPS1 patients to the controls with complaints, whereas changes in sudomotor/ edema and motor/trophic signs were more alike.
Table 3 presents data on differences in skin surface temperature between the involved and noninvolved side as calculated by the various mathematical methods for each of the three groups. There is an overall significant difference comparing all three groups for all measurements except absolute static temperature difference between wrist/ankle and fingertips/toe tips. The difference between CRPS1 patients and control patients with complaints is significant for Euclidian distance and total of difference between fingers/toes. There was a significant difference found between CRPS1 patients and control patients without complaints in the absolute difference in mean hand/foot temperature, in asymmetry factor, in Euclidian distance. The difference in temperature between the involved and noninvolved side (as indicated by the various calculation methods) shows a tendency to decrease; the largest difference in temperature was found in CRPS1 patients compared with the two other groups.
The differences in temperature between the palmar/ plantar side and dorsal side are small and not significant; therefore only the palmar side of hands and feet are presented in Tables 3 and 4.
Indicators which reflect the diagnostic value (such as sensitivity and specificity) are presented in Table 4. The positive predictive value, negative predictive value, and AUC are similar for all calculation methods, whereas the sensitivity and specificity differ. The absolute difference in mean hand/foot temperature was a weak predictor of CRPS1 patients, whereas average fingertip temperature, asymmetry factor, Euclidian distance, and total difference in temperature between fingers and toes proved to have stronger diagnostic value.
Significant differences between CRPS1 patients, control patients with complaints, and control patients without complaints were found, as reflected by the various mathematical methods used to calculate temperature differences. ROC analysis of the diagnostic value of thermography revealed a moderate discriminating power, as indicated by an AUC of ≤0.7. Furthermore, not every mathematical method showed a significant difference among the three subgroups. The control patients with complaints showed overlap with CRPS1 patients in both symptoms and signs. Moreover, there was a large overlap between symptoms; most prominent in CRPS1 fracture patients were the displayed vasomotor symptoms, indicating that vasomotor signs belong to the most prominent signs at early onset of CRPS1; this has also been reported by others.1,6,8,10–14 The difference in temperature between the involved and noninvolved side (as indicated by the various calculation methods) shows a tendency to decrease, the highest difference expressed by the CRPS1 patients and the lowest difference observed in control patients without complaints; this phenomenon was also found in studies by Schurmann et al. and Bruehl et al.8,10,12
Studies on the use of thermography to discriminate between CRPS1 patients and non-CRPS patients lack consistency regarding the calculation methods used on the thermographic data (e.g., whole-hand calculations, spots, and fingertips), the statistical analyses used, as well as the description of and inclusion criteria applied. Birklein et al.6 were unable to calculate specificity and negative/positive predictive values because they did not include patients with symptoms similar to CRPS1. Gradl and Schurmann11 used an arbitrary fixed cutoff value of >1.5°C for fingertip/hand temperature difference and found that thermography had a low sensitivity and specificity; moreover, they did not report on positive/negative predictive values, nor did they use ROC analysis to derive indices on diagnostic value at other cutoff values. Shurmann et al. used only fingertip temperature to calculate the difference between the involved and noninvolved extremity and did not compare other calculation methods on thermographic data.12 There is some overlap of the absolute mean temperature difference between the involved and noninvolved extremity found in CRPS patients between the present study and values reported in the literature. ROC analysis results in a range of cutoff values with an associated sensitivity and specificity.
In this study, we found a difference of >0.99°C on absolute mean temperature difference between extremities to be the optimum cutoff value as indicated by ROC analysis, whereas Bruehl et al.10 found a difference of >0.82°C to be optimum. Furthermore, in the present study a mean asymmetry factor of 0.39 was found for patients and 0.83 for control patients without complaints. In our previous study, a mean asymmetry of 0.45 was found for CRPS1 patients and 0.91 for healthy controls.5 In the literature, sensitivity is reported to range from 58% to 93% and specificity from 86% to 89%8,11,15; none of these latter studies reported on AUC. In our study, the highest sensitivity was 71% and the highest specificity 64%, with a moderate AUC of >0.63. In this study, the cutoff value was chosen that resulted in the highest combination of sensitivity and specificity. A different cutoff value could also have been considered; however, because of the low AUC as calculated by the ROC, a different cutoff value will only alter the sensitivity at the cost of specificity. In conclusion, a different cutoff does not improve the diagnostic capabilities as a whole. The positive and negative predictive values can be calculated using the calculated sensitivity and specificity combined with the incidence of the studied population. In this study, fracture patients with a high risk of CRPS1 were included; therefore in the analysis, the incidence of the studied population was used. Only Gulevich et al.15 reported on positive predictive value (90%) and negative predictive value (94%). In our study, the highest positive predictive value was 35% and the highest negative predictive value was 84%. The reason for the high level of positive predictive value found in the study of Gulevich et al. compared with this study could be because of the sympathetic stress methods used in that study.
Although we were only interested in the diagnostic value of temperature recordings at the very early stage after the onset of CRPS1, one of the limitations of this study is a lack of follow-up. Data obtained after 6, 12, and 18 wk could emphasize the difference between CRPS1 patients and slow recovering control patients with complaints. None of the patients with pain but without CRPS1 developed CRPS1, according to the second follow-up visit 8 wk later.
One could argue that temperature difference in CRPS1 patients could already have spread outside the fractured area, thus increasing the difference between the CRPS1 group and the control fracture group. However, the difference between the involved and noninvolved leg was not significant in any of the three groups, nor was this difference significant among the three groups.
The validity of thermographic recording to discriminate between CRPS fracture patients and control patients with complaints at the early onset of CRPS1 is limited; therefore, thermography should be considered as an additive diagnostic tool. When an abnormal pattern of temperature is identified and confirmed using the cutoff values, as indicated in the results of this study, a follow-up of these patients is advisable. In that case, the best performing mathematical methods that are able to evaluate all the collected thermographic data are the asymmetry factor, the Euclidian distance, and total of variation between fingers and toes. There is no preference for recording on the palmar/plantar or dorsal side other than for pragmatic reasons.
Some studies have shown an increase in temperature difference after methods in which the sympathetic system is provoked.14,16 Further research should focus on this finding and investigate the effect on temperature difference between extremities and the effect on the various indicators for diagnostic purposes.
This study was performed within TREND (Trauma Related Neuronal Dysfunction), a knowledge consortium that integrates research on Complex Regional Pain Syndrome type 1. The project is supported by a Dutch Government grant (BSIK03016). The authors thank Prof. Rianne de Wit for her assistance in the setup of this study and Laraine Visser-Isles (Department of Anesthesiology) for assistance with the English language.
1. Bruehl S, Harden RN, Galer BS, Saltz S, Bertram M, Backonja M, Gayles R, Rudin N, Bhugra MK, Stanton-Hicks M. External validation of IASP diagnostic criteria for complex regional pain syndrome and proposed research diagnostic criteria. International Association for the Study of Pain. Pain 1999;81:147–54
2. Quisel A, Gill JM, Witherell P. Complex regional pain syndrome underdiagnosed. J Fam Pract 2005;54:524–32
3. Janig W, Baron R. Complex regional pain syndrome is a disease of the central nervous system. Clin Auton Res 2002;12:150–64
4. Baron R, Janig W. Complex regional pain syndromes—how do we escape the diagnostic trap? Lancet 2004;364:1739–41
5. Huygen FJ, Niehof S, Klein J, Zijlstra FJ. Computer-assisted skin videothermography is a highly sensitive quality tool in the diagnosis and monitoring of complex regional pain syndrome type I. Eur J Appl Physiol 2004;91:516–24
6. Birklein F, Riedl B, Neundorfer, Handwerker HO. Sympathetic vasoconstrictor reflex pattern in patients with complex regional pain syndrome. Pain 1998;75:93–100
7. Gradl G, Steinborn M, Wizgall I, Mittlmeier T, Schurmann M. [Acute CRPS I (morbus sudeck) following distal radial fractures–methods for early diagnosis]. Zentralbl Chir 2003;128:1020–6
8. Schurmann M, Gradl G, Andress HJ, Furst H, Schildberg FW. Assessment of peripheral sympathetic nervous function for diagnosing early post-traumatic complex regional pain syndrome type I. Pain 1999;80:149–59
9. Herry CL, Frize M. Quantitative assessment of pain-related thermal dysfunction through clinical digital infrared thermal imaging. Biomed Engl Online 2004;3:19
10. Bruehl S, Lubenow TR, Nath H, Ivankovich O. Validation of thermography in the diagnosis of reflex sympathetic dystrophy. Clin J Pain 1996;12:316–25
11. Gradl G, Schurmann M. Sympathetic dysfunction as a temporary phenomenon in acute posttraumatic CRPS I. Clin Auton Res 2005;15:29–34
12. Schurmann M, Gradl G, Zaspel J, Kayser M, Lohr P, Andress HJ. Peripheral sympathetic function as a predictor of complex regional pain syndrome type I (CRPS I) in patients with radial fracture. Auton Neurosci 2000;86:127–34
13. Wasner G, Schattschneider J, Baron R. Skin temperature side differences–a diagnostic tool for CRPS? Pain 2002;98:19–26
14. Wasner G, Schattschneider J, Heckmann K, Maier C, Baron R. Vascular abnormalities in reflex sympathetic dystrophy (CRPS I): mechanisms and diagnostic value. Brain 2001;124:587–99
15. Gulevich SJ, Conwell TD, Lane J, Lockwood B, Schwettmann RS, Rosenberg N, Goldman LB. Stress infrared telethermography is useful in the diagnosis of complex regional pain syndrome, type I (formerly reflex sympathetic dystrophy). Clin J Pain 1997;13:50–9
16. Niehof SP, Huygen FJ, van der Weerd RW, Westra M, Zijlstra FJ. Thermography imaging during static and controlled thermoregulation in complex regional pain syndrome type 1: diagnostic value and involvement of the central sympathetic system. Biomed Engl Online 2006;5:30