Link to Figures Show Figures
	Article Text Figures References
	Send Link Send HTML Article

Measurement of Socket Discomfort--Part II: Signal Detection

Edward S. Neumann, P.E, PhD

The fitting of sockets requires discriminative judgments by both the prosthetist and the patient. If the socket is too tight or creates excessive pressures at specific locations, discomfort and possibly tissue trauma may result. The detection of an unsatisfactory fit may in some cases occur quickly and without uncertainty. However, in other situations, uncertainty may be involved. Patients may experience sensations that are not conclusive in terms of whether or not discomfort will occur after the initial fitting and must attempt to form judgments. Usually, it takes days or weeks for a patient to settle into a new socket. A socket that fits well initially may become uncomfortable, and it can be difficult for the patient to predict this.

Signal detection theory (SDT) offers a psychophysical model of discrimination which can be applied when there are two discrete states of the world possible--either just noise is present or a signal is present--and it is not easy to discriminate between the two states. ^1,2 Examples of such discriminative situations include the reading of radiograph plates to detect tumors, the reading of polygraphs, and the interpretation of aerial photos.

SDT has been used to study recognition memory and eyewitness testimony, the effect of placebos and drugs on pain, how clinicians form judgments on the presence or absence of disease, the vigilance of individu1als monitoring aircraft surveillance radars, and manufacturing inspection. Signal detection judgments involving uncertainty are not simple but involve a number of factors. SDT incorporates several key variables that facilitate an understanding of the judgment process. The goal of this study was to examine the fitting process from a SDT perspective and design an experiment to produce data that could be used to evaluate the feasibility of calibrating SDT models of socket fit. SDT models would provide researchers with an additional tool for evaluating socket materials and designs, and could have clinical applications.

SDT MODEL OF SOCKET FITTING

SDT is applicable to situations in which an individual must discriminate between two neurological states--a noise-alone background state and a background noise plus signal state. In the context of socket fit, it can be used to model the ability of a patient to detect whether the sensations resulting from a specific socket geometry will cause discomfort. ( Figure 1 ) illustrates the key concepts. The noise-alone distribution is to the left, and the noise-plus-signal distribution is to the right. The criterion is the level of sensation above which the subject perceives and reports a signal to be present, and below which the subject perceives and reports no signal to be present. If the two distributions overlap, then a sensation above the criterion has a finite probability of either containing a signal or being just noise. Similarly, a sensation below the criterion has a finite probability of either containing a signal or being just noise. The probability that a subject will correctly identify a signal when it occurs depends on where the criterion is set. As the criterion is moved toward the left, the probability that a signal will be detected when it occurs increases (a hit) but so does the probability that the subject will say a signal is present when it is not (a false alarm). As the criterion is shifted toward the right, the probability that the subject will report no signal when none has been presented increases (correct rejection) but the probability a signal will be missed when one is presented also increases (a miss). Research on signal detection has demonstrated that the criterion set by a subject is not an absolute sensory threshold but varies according to the relative costs and benefits of the four possible outcomes (hit, miss, correct rejection, and false alarm). If SDT is applied in the context of the socket-fitting process, the background state can be defined as the perceived pressure between the socket interface and the residual limb, and the signal can be defined as perceived discomfort or the perception that pressure is sufficiently high to produce long-term discomfort and possibly pain. When a patient is asked if discomfort is felt or expected at a particular location in the socket (such as the distal end of the tibia), there are four possible outcomes, as follows: 1. The pressure is high enough to cause discomfort, pain, or tissue breakdown after settling in, and the patient reports that he or she expects the socket to cause discomfort or pain. In SDT, this is termed a "hit." 2. The pressure is high enough to cause discomfort, pain, or tissue breakdown after settling in, but the patient reports that he or she does not expect the socket to cause discomfort or pain. This is termed a "miss." ³ . The pressure is not high enough to cause discomfort, pain, or tissue breakdown after settling in, and the patient reports that he or she does not expect the socket to cause discomfort or pain. In SDT, this is termed a "correct rejection." ⁴ . The pressure is not high enough to cause discomfort, pain or tissue breakdown after settling in, but the patient reports that he or she expects the socket will cause discomfort or pain. This is termed a "false alarm."

Perfect patient performance would consist of only hits and correct rejections. Misses result in later discomfort or tissue breakdown problems for the patient, and false alarms result in unnecessary modifications to the socket. SDT states that the probabilities of the four outcomes are a function of the response bias and sensitivity of the patient, as well as the strength of the signal and the expectation that the signal will occur. Response bias refers to the tendency of the patient to favor one response over another, and the two extremes of response may be defined as risky or conservative biases. According to SDT, patients who wish to maximize the number of hits have a risky bias and can accomplish this by setting the subjective criterion for signal detection very low, which ensures that among those locations identified within the socket will be included as many as possible of those which the patient thinks might cause discomfort. However, the cost for doing this could be a high number of false alarms because the individual with a risky bias also will tend to report expectations that specific locations in the socket will cause discomfort or pain when this actually might not occur once the patient gets used to the socket. Patients who have a conservative bias do just the opposite; they seek to minimize the number of false alarms. In the process, they set a high criterion for discomfort and wind up underestimating the number of socket locations that actually will cause discomfort or tissue breakdown. Thus, they tend to develop expectations during fitting that socket locations where discomfort or pain is being experienced may not cause discomfort or pain after the socket has been worn for several days (misses). SDT states that just where along the risky-conservative continuum a patient sets his or her criterion is determined by the expected payoff of the outcomes, Bopt, which is expressed mathematically as Bopt=Prob (Noise)×(benefit of correct rejection+cost of false alarm)Prob (Signal)×(benefit of hit+cost of miss)

False alarms and misses result in costs to the patient, whereas correct rejections and hits produce benefits. From the perspective of the patient, factors influencing the cost of a miss could include, among other considerations: 1) medical problems resulting from tissue breakdown, 2) functional level of the patient and the impact of loss of mobility or ability to wear and use a prosthesis, ³ ) the pain and discomfort tolerance level of the patient, and 4) the resources required and problems involved in scheduling and making a return visit to the prosthetist. Factors that could influence the cost of a false alarm are: 1) reduction in overall quality of socket fit following geometric adjustment in a specific location, 2) loss of socket structural integrity, ³ ) time availability of the patient or the physical endurance required of the patient to wait while adjustments are made to the socket, and 4) the perception by the patient that the time availability of the prosthetist is limited.

The SDT concept of Bopt demonstrates a possible link between the technical and interpersonal skills of the prosthetist and the signal detection performance of the patient. If the prosthetist becomes irritated or alienated when socket fit is questioned, this could increase the cost of a false alarm (as well as create a cost for a hit!). Conversely, if the prosthetist appears to welcome the opportunity to resolve discomfort problems, this could increase the perceived benefit of a hit and reduce the perceived cost of a false alarm. If a patient believes that a prosthesis is inherently supposed to be uncomfortable or cause pain, this could also could influence Bopt; the benefit of a hit might be perceived to be low if the patient believes that nothing can be done to reduce discomfort.

Misses can be especially costly to the patient if they result in pain or tissue breakdown, and hits are very beneficial because they may help prevent tissue breakdown or pain. False alarms may have little cost unless they result in socket alterations that reduce the overall quality of the fit or the integrity of the socket or require the patient to spend a longer time in the prosthetics clinic than desired. Correct rejections imply a good fit, and no need for modification of the socket. The lower the value of Bopt that a patient adopts (risky bias), the lower the number of misses the patient is willing to accept, and the greater the number of false alarms he or she is willing to produce. If the perceived cost of a miss is high to a patient and the penalty of a false alarm is low, the patient may tend to have a risky bias. This could be the case, for example, if the patient were facing a high cost due to loss of mobility should he or she experience pain or tissue breakdown and the prosthetist were very willing to make socket adjustments. If it were difficult for a patient to return to the prosthetist for discomfort adjustments, a risky bias might be appropriate.

The higher the value of Bopt that the patient adopts (conservative bias), the fewer the number of false alarms that he or she desires to occur, and the greater the number of misses he or she is willing to accept. If the perceived cost of a miss is low and the penalty for a false alarm is high, a patient might tend to have a conservative bias. This bias could occur, for example, if a patient did not expect to wear the prosthesis much and the prosthetist appeared unwilling to make socket adjustments. Or, if it were very easy for the patient to return to the laboratory for a discomfort adjustment, the cost of a miss could be low. Alternatively, if a patient did not like to complain (the psychological cost of incorrectly identifying a problem area in the socket is high) or thought that a painful or uncomfortable fit was normal (the value of a hit is low because nothing can be done about discomfort), he or she might adopt a conservative bias. Patients who are concerned that unnecessary alternations to the socket could worsen fit also might tend to adopt a conservative bias, which could result in frequent return visits for adjustments.

The expectation or probability that a signal will occur also influences response bias. According to SDT, if a patient expects that a signal is highly probable, a more risky bias will tend to be adopted. Thus, a patient who has frequently experienced discomfort problems in the past during the fitting process but had them successfully resolved by the prosthetist may be more likely to set a low criterion along the continuum, and a patient who has seldom experienced discomfort during fitting, or has not had success with attempts to eliminate discomfort, could be more likely to set a high criterion.

Sensitivity concerns the ease with which the signal can be detected. It is measured by the distance between the means of the two distributions along the subjective continuum. If the distributions of noise-alone and noise-plus-signal are widely separated on the psychophysical continuum and there is little overlap of the distributions, then the proportion of hits and correct rejections will be high. This might be the case for experienced wearers with good sensation. If the means of the two distributions are close together and the distributions have considerable areas of overlap, then the patient will have trouble distinguishing noise from signal, and a larger proportion of false alarms and misses will occur. This could be the case for individuals with poor sensation and possibly inexperienced or first time wearers.

The results obtained from applying SDT in an experiment often are displayed by means of a receiver-operating characteristic curve (ROC), which plots the false alarm rate versus the hit rate for a given pair of distributions representing noise-alone and noise-plus-signal when the response bias is varied through experimental means. When sensitivity is zero (no difference between the means of the two distributions along the psychological continuum), the plot approximates a straight line oriented at 45° from the origin. As sensitivity increases and signal detection performance improves, the plot forms a progressively greater convex curve that approaches the upper left corner of the graph.

Standard experimental procedures for generating an ROC curve involve presenting both noise-alone and noise-plus-signal stimuli several hundred times under varying, experimentally induced response biases and calculating the probabilities associated with all four outcomes. However, this would be infeasible in a clinical setting and most prosthetic research settings. An alternative procedure, which would be appropriate for clinical and research use, involves a small number of trials combined with asking the patient to provide an estimate of his or her subjective uncertainty about the presence or absence of a signal. The frequency with which uncertainty estimates fall into categories ranging from "very certain" to "very uncertain" can substitute for experimental variation of the response criterion.

EXPERIMENTAL METHODS

Three subjects with unilateral transtibial amputations who had been fitted with PTB sockets and Pelite liners were chosen from the patient clientele at the residency site and invited to participate in the experiment. All subjects were volunteers, and each read and signed a letter of informed consent that had been approved along with the Human Subject Protocol by the Biomedical Sciences Committee of the Institutional Review Board of the University of Nevada, Las Vegas. Two of the subjects (subjects 1 and 3) experienced war-related traumatic amputations in 1944 and 1968 and had worn a prosthesis for many years, but the third subject (subject 2) was in the process of being fitted for a first prosthesis after amputation surgery seven months earlier and thus had no prior experience with use of a prosthesis. All of the subjects had touch thresholds below 10-g force at the distal tibia and fibula head, and none had impaired spatial acuity of touch; thus, all appeared to have normal levels of sensation in their residual limbs. Additional information on the subjects is presented in Table 1 .

To duplicate each subject's fitted socket as closely as possible, plaster was poured into the subject's Pelite liners, and a Duraplex check socket was bubble-formed over the resulting models. The number of ply of sock that the subject currently was wearing was pulled over the residual limb prior to donning the Duraplex socket, and each subject was asked to verify that the fit was similar to their definitive prosthesis. Minor relief adjustments were made in the sockets, if needed.

Flexible copolymer wafers approximately 1 mm thick and 23 mm in diameter were taped inside the Duraplex sockets at the distal tibia and the apex of the fibula head, which produced variations in the socket geometries and pressures experienced by the subjects. Zero, one, or two wafers were inserted randomly at each location such that each of the three possible states (0, 1, or 2 wafer thicknesses) was presented three times at each site, for a total of nine combinations of geometries. It was assumed before the experiment that the 0 wafer condition would represent a noise-alone state and the 1 and 2 wafer conditions would represent two different noise-plus-signal states, which would enable the calibration of two different signal detection models per subject. In an attempt to correlate the responses of the subjects to internal socket pressures, a low-cost socket pressure sensor also was taped over the wafers and directly on the socket surface for the no-wafer condition. The thickness of the sensor was 0.5 mm, and the thickness of the tape used to hold the wafers and sensor in place was 0.1 mm, making the total thicknesses of the three conditions 0.6, 1.6, and 2.6 mm.

Before the experiment, it was expected that the 0.6-mm thickness of the sensor and tape alone (zero wafer condition) would be undetectable by the subjects. During data collection, this assumption was found to be incorrect for subject 1, as will be discussed below. Also, it was found that the pressure sensor data were not reliable because the curvature of the socket produced spurious pressure readings independently of the force being applied by the residual limbs. Thus, the pressure data were not incorporated into the analyses.

For each of the nine conditions, subjects donned their socks and check sockets and stood with a height-adjustable platform supporting their residual limb. The platform height was adjusted to equalize hip height, and subjects were asked to place their body weight on the residual limb and rock back and forth in their check sockets. These procedures were the same as those used during routine fitting of a PTB check socket in the clinic. Additional tasks for the experiment involved asking the subjects, for each of the nine experimental conditions, to form judgments on the subjective magnitudes of pressure, discomfort, and pain being experienced at each site. A Borg category ratio scale was utilized for responses. The judgment tasks and the response categories presented to the subjects are shown in ( Figure 2 ). To obtain data for the analysis of signal detection performance, the following question was asked:

f If no changes were made to the socket at this location, how do you think it would feel after several days of normal use? Please select the answer that best describes what you would expect. 1. VERY SURE it would cause discomfort or pain 2. FAIRLY SURE it would cause discomfort or pain ³ . NOT SURE if it would cause discomfort or pain ⁴ . FAIRLY SURE it would NOT cause discomfort or pain ⁵ . VERY SURE it would NOT cause discomfort or pain.

This question sought to measure the subjective uncertainty associated with the judgments and enabled the signal detection models to be calibrated with a small number of data points using SYSTAT 8.0. Because there were five categories of response and only three observations for the noise-alone condition and three observations for the noise-plus-signal condition at each site, the data set was underdetermined for the statistical calibration procedure. To circumvent this problem, the frequencies for each observation were multiplied by a factor of 10, and one additional observation was generated for any category that had received zero responses. This ensured that each of the five categories was represented at least once for both the noise-alone and noise-plus-signal conditions. The multiplication of each response by a factor of 10 did not introduce any error because it was applied uniformly to all observations. The creation of one additional data point, when needed to ensure that each category was represented at least once, affected less than 10% of the total observations used to calibrate the models, which minimized the impact on calibration. However, the calibration procedure assumed that both the noise-alone and noise-plus-signal distributions were normal. Expansion of observed frequencies by a factor of 10 in combination with the creation of single observations for unreported categories may have accentuated any non-normal properties of the responses, which would effect goodness-of-fit measures. Given that the endurance of the subjects for repeated donning and doffing of the check sockets was limited, and the potential for judgment fatigue was high, no option was feasible other than expansion of the reported responses and creation of a limited number of additional data points for purposes of calibration.

RESULTS AND DISCUSSION

For all experimental conditions of noise-alone and noise-plus-signal, subject 1, who was an experienced prosthesis user, reported the same values for pressure perception, discomfort, and expectation at the distal tibia and fibula head. Pressure sensation was reported as "moderate," discomfort was reported as "mild," and expectation was reported as "VERY SURE it [the socket] would cause discomfort or pain." These values were given even for the conditions when no wafers were inserted in the socket, but only the 0.6-mm pressure-sensing element and tape were present. This result was not expected before the experiment. However, subject 1 self-reported being a very difficult patient to fit, who approached socket comfort modifications with trepidation, because very minor changes in local geometry could cascade throughout the entire socket and cause pressure redistributions that were uncomfortable. Subject 1 may have learned through experience and was able to detect that the increased level of pressure, which produced only "mild" discomfort, would make the socket intolerable during extended use. Whereas the other two subjects reported variations in pressure sensation and discomfort with different wafer thicknesses, subject 1 did not. Reasons for this were not clear but may have been related to other anomalous responses provided by subject 1 during related experiments, which are reported elsewhere. ⁴ Because of the lack of variation in response for subject 1, a signal detection model could not be calibrated.

Figure 3 presents for subjects 2 and 3 the relationships between insert thickness and the expectation that the socket geometry would cause discomfort or pain. The lower the number on the expectation axis, the greater the certainty that discomfort or pain will result. Although most of the data varied in the manner expected, only one of the correlations was significantly different from zero at the 0.05 level (subject 3 at the fibula head). The lack of significant correlations and variation in expectation response for each of the conditions indicated that the subjects were facing uncertainty with respect to signal detection. The variation could have been due to variations in judgment from trial to trial, which is the basis for calibrating a signal detection model, or due to small variations in positioning of the wafers and pressure sensor, and the donning of the socket. As will be discussed below, the signal detection model provided additional information on the subjects' performance not obtainable simply from an analysis of correlations. For the experimental condition of zero inserts in the socket but just the pressure sensor, 75% of the time subjects reported expectations of "NOT SURE," "FAIRLY SURE," or "VERY SURE" that the socket would not cause discomfort or pain, whereas with two wafers in the socket, 83% of the time subjects reported expectations of "FAIRLY SURE" or "VERY SURE" that the socket geometry would cause pain or discomfort.

Figure 4 plots the relationship between pressure perception and expectation. The R-square values were all significantly different from zero at the 0.05 level. For subject 2, who was inexperienced, the slopes of the lines for both the distal tibia and the fibula head were numerically close, though the R-square value was higher for the distal tibia, which was a pressure sensitive area for this subject during fitting. For subject 3, who was experienced, the slope for the fibula head was steeper than the slope for the distal tibia, though the correlation was lower. Subject 3 had reported the fibula head as a troublesome site during socket fitting. Comparing the data points and slopes obtained for subject 2, who was inexperienced, with those for subject 3, who was experienced, it appeared that the more experienced subject 3 tended to make fewer "NOT SURE" judgments and more "VERY SURE" judgments.

The relationship between magnitude of discomfort and expectation is plotted in ( Figure 5 ). Most of the correlations were higher than those for the relationship between pressure perception and expectation. This difference in correlation magnitudes suggests a model of expectation in which pressure sensation information first is subjectively weighted by one or more additional criteria to arrive at an immediate discomfort judgment (a negative valence is assigned to the pressure sensation), and the discomfort judgment is then evaluated subjectively to arrive at a discomfort prediction or expectation. The expectation judgment would be influenced by prior experience, which also could play a role in the assignment of a negative valence to a pressure sensation. A feature of this hypothesis is that the immediate discomfort sensation would represent an important intervening variable between pressure sensation and discomfort expectation, and it would be important to measure and account for it in studies of discomfort. The implication is that pressure sensation alone would not be an accurate predictor of expected discomfort. However, further research would be necessary to verify this model.

For each subject, the regression coefficients for both the distal tibia and fibula head were numerically very similar, though they differed between the two subjects (-0.50 and -0.48 for subject 2; -0.79 and -0.77 for subject 3). The closeness of the within-subject coefficients may be an indication that the relationship between the perceived magnitude of discomfort and expectation for each subject was consistent over the two different sites within the socket, which would suggest similar underlying sensory response functions and signal detection criteria within subjects but variation between subjects. R-square values for all four curves were the highest of any of the relationships between the psychophysical scales and expectation, indicating a good correspondence perceived discomfort and expectation. The steeper slopes for subject 3 imply that in comparison with subject 2, subject 3 expected areas of the socket that were perceived as creating "no" or "just noticeable" discomfort to have a slightly greater likelihood of not causing problems and areas that were perceived as creating "somewhat strong" discomfort were slightly more likely to cause problems. This may have reflected subject 3's much longer experience with prosthesis use.

Measures of ROC fit include MN, the mean of the noise-alone distribution, which is set equal to zero for calibration; SDN, the standard deviation of the noise-alone distribution, which is set equal to 1.00; MN+S, the mean of the noise-plus-signal distribution; SDN+S, the standard deviation of the noise-plus-signal distribution; D, the sensitivity of the subject, or the distance between the means of the noise-alone and the signal-plus-noise distributions; the ROC area, which measures the proportion of the total area that lies under the ROC curve and indicates the extent to which the performance of the individual is better than chance; and the chi-square probability for goodness of fit. An ROC area of 0.50 indicates that results are pure chance, whereas an ROC of 1.00 indicates perfect performance; numbers in between 0.50 and 1.00 indicate the extent to which performance is better than chance but less than perfect. A chi-square probability of 1.00 represent a perfect fit to the assumption of normality; as chi-square probability approaches 0.00, fit is increasingly poor.

Examination of the hit versus false alarm rates for each subject contrasts their differences in terms of signal detection performance. For the 2.6-mm insert at the distal tibia, subject 2 would be predicted to have a false alarm rate of only 10% when this individual's criterion was set to produce a hit rate of 60%. But if subject 2's criterion was shifted to produce a higher hit rate of 90%, the predicted false alarm rate would be 50%. For subject 3, a hit rate of 60% would be associated with a false alarm rate of less than 5%, and a shift in the criterion that produced a 90% hit rate would have predicted false alarm rate of only 20%. Subject 3's false alarm rate was less than half the false alarm rate of subject 2 for the experimental conditions.

The resulting ROC plots and measures of fit are presented in ( Figure 6 ) and ( Figure 7 ) for subjects 2 and 3, respectively. The ROC plot indicates the relationship between hit rate and false alarm rate for the subject. The closer the curve is to the 45° line running from the lower left corner to the upper right corner, the closer the noise-plus-signal distribution lies to the noise distribution, and the more guessing the subject is undertaking because signal detection ability is poor. The more convex the curvature, the less guessing that the subject is undertaking, and the greater discrimination ability is. The circles show the locations of the upper boundaries of each of the five categories. As calibrated, the shape of the curves reflects the assumption of normal distributions. Were the normality assumption to be dropped and a nonparametric fit attempted, the curves would consist of straight lines connecting the circles representing the upper category boundaries. However, no measures other than the area under the ROC curve would be available. The

ROC plots for subject 2 indicated that discomfort expectations with 1.6-mm inserts at both the distal tibia and the fibula were no different from discomfort expectations when only the 0.6-mm pressure sensor was present. MN+S values for the 1.6-mm inserts were actually slightly to the left of the MN values for the 0.6-mm pressure sensor. The areas under the ROC plots (0.426 and 0.406) were lower than those for pure guessing. Hit rates were lower than false alarm rates and signal detection ability was quite limited. With the 2.6-mm insert, signal detection improved, and the MN+S values were 1.721 and 1.010 standard deviations to the right of the MN values, indicating sensitivity was greater at the distal tibia than at the fibula head. The areas under the ROC plots (0.847 and 0.702) for the 2.6-mm inserts indicated performance nearly halfway and more than halfway between pure guessing and perfect signal detection. The plots show that hit rates were greater than false alarm rates, with signal detection for the 2.6-mm insert at the distal tibia being somewhat better than signal detection for the same insert at the fibula head. However, the results at the fibula head might also have been influenced by the thick tissues of the subject in this area, as well as the possibility that the socket was not fitting tightly, both of which could have made detection more difficult. Also, the distal tibia had been a sensitive location during fitting. The SDN+S values were greater than 1.00 for all plots, indicating a greater variation in response for the noise-plus-signal condition than for the noise-alone condition. These results might reflect subject 2's lack of prior experience wearing a socket. Goodness of fit values reveal that only the 2.6-mm insert at the fibula head produced responses that approximated well the normality assumptions underlying the model.

The ROC plots for subject 3 indicated better signal detection than subject 2 at both the distal tibia and fibula head. Subject 3 also reported various level of pain at the fibula head during the experiment, so good performance was expected. MN+S values were more than one-half and more than one standard deviation to the right of MN values for 1.6-mm inserts at the distal tibia and fibula head, respectively, and well to right of MN values for the 2.6-mm inserts--1.429 and 8.44 standard deviations. This can be interpreted as a very high sensitivity to pressure and extremely good signal detection at the fibula head, a location that subject 3 self-reported as experiencing fitting difficulties in the past. It is likely that the perception of pain heightened awareness of a potential fitting problem and had an influence on expectation responses. The high values of the ROC areas indicate that subject 3's signal detection ability for the 1.6-mm inserts was comparable to subject 2's ability for 2.6-mm inserts, and subject 3's signal detection ability for the 2.6-mm insert at the fibula head was almost perfect. The SDN+S values at the distal tibia were less than 1.00, indicating a smaller variation in response for the noise-plus-signal condition than for the noise-alone condition. At the fibula head, SDN+S values were greater than 1.00. Although SDN+S was quite high for the 2.6-mm insert at the fibula head (5.602), it was still much less than MN+S (8.442), indicating that the two distributions had little overlap. Likewise, there was good separation between the two distributions at the distal tibia for the 2.6-mm insert, though the distributions had more overlap than at the fibula head. The ROC plots reveal very high hit rates and very low false alarm rates for the 2.6-mm inserts, and good hit rates and low false alarm rates for the 1.6-mm inserts. This could reflect the thinness of the compressible tissues on subject 3's residual limb and a snugly fitting socket, but it also would be consistent with the hypothesis that signal detection improves with wearing and fitting experience. Chi-square goodness of fit measures indicated good model fits for the 1.6-mm insert at the distal tibia and the 2.6-mm insert at the fibula head, and a poor fit for the other two plots, which was a consequence of having a small number of experimental data points.

CONCLUSIONS

The terms comfort and discomfort frequently are used when discussing the problems of socket design and fit. Although it is tempting to consider them as anchors at the opposite ends of a psychological continuum, the neurological bases of each may have noteworthy differences. Damasio states that "Pain and pleasure are not twins or mirror images of each other . . . there seem to be far more varieties of negative than positive emotions, and it is apparent that the brain handles positive and negative emotions with different systems." ⁵ The science of neurophysiology has not made as much progress in the understanding of pleasure-related phenomena as it has in the understanding of neural responses necessary for survival. For this reason, an understanding of discomfort is the appropriate starting place to begin the quest for comfort. To achieve comfort, the first task is to eliminate painful and distressing experiences by managing the stimuli that cause discomfort. There can be no comfort when discomfort is present.

SDT entails a number of concepts that are relevant to understanding discomfort perceptions and socket fitting. One of its most significant contributions is the insight that the socket discomfort judgments made by patients may not always be simple responses to perceived sensations but actually could be complex decisions that are influenced by the relative costs and benefits of hits, correct rejections, false alarms, and misses. Further, the attitudes and interpersonal skills of the prosthetist could influence some of these perceived benefits and costs. Despite several experimental design problems, including the ability of one subject to detect a thin pressure sensing pad and judge it as an unacceptable modification to socket geometry that would cause discomfort, and difficulty reproducing exactly the wafer locations and pressure sensations during each trial, the research demonstrated that it is possible to design an experiment and collect data that can be used to calibrate a signal detection model that quantifies the ability of a patient to predict the longer term quality of socket fit and indicates the patient's hit rate versus false alarm rate as their decision criterion shifts. Thus, SDT is a tool that could be used for research on socket design and materials. It is recommended that if the procedures of this study are used, a larger number of trials be conducted to have more data points for model calibration.

SDT also could play a role in the clinic. An SDT approach to fitting would enable the prosthetist to focus on the basic problem of short-term socket fit as might be achieved during a single fitting session versus longer-term fit--how the socket will feel the next day or a week later. It is not unusual for a patient to leave a fitting session with apparent satisfaction, only to find that discomfort develops after several hours or days of use. The SDT-relevant question developed for this research asks the patient to estimate how certain they are that the socket will not cause discomfort or pain after several days of normal use. If the patient has some degree of uncertainty concerning how the socket will feel, the question gives them an opportunity to express this uncertainty to the prosthetist and alerts the prosthetist to a possible future problem.

The data produced by a calibrated SDT model provide insight into a number of psychophysical parameters that may vary among patients and help explain why some patients present more of a challenge when trying to eliminate socket discomfort than others. Sensitivity is one of the key parameters. The greater the distance along a psychological continuum between the means of the noise-alone and noise-plus-signal distributions, the better signal detection will be performed. Conversely, the lower the sensitivity, the more poorly signals will be detected. The neuropathic individual who has no sensation in their residual limb, and thus no sensitivity will, obviously, be unable to detect signals.

Two additional variables not addressed by this study merit consideration in future research. The first variable is the duration of wear needed for an individual to evaluate socket discomfort. Research on seating comfort indicates time periods ranging from 5 to 30 minutes may be needed to evaluate seats.6 The measurements for the nine conditions in this study took approximately an hour per subject, but the sockets were donned for only several minutes during each condition. The second variable concerns evaluation during static versus dynamic fitting. The present study examined only static fitting. Dynamic fitting, which would involve having the subject walk on an aligned prosthesis, would more accurately reproduce the forces and pressures encountered in every day wear.

ACKNOWLEDGMENTS

The author is indebted to Jerry Fullerton, CPO, and owner of Superior Limb and Brace, Las Vegas, for providing support and encouragement during the project. Note: This research was self-financed by the author to fulfill his Prosthetics Residency research project requirement.

References:

1. Wickens CD, Hollands JG. Engineering Psychology and Human Performance, 3rd ed. Upper Saddle River, NJ: Prentice Hall; 2000:17-44.
2. Gescheider GA. Psychophysics, 3rd ed. Mahwah, NJ: Lawrence Erlbaum; 1997:105-182.
3. Borg G. Borg's Perceived Pain and Exertion Scales. Champaign, IL: Human Kinetics; 1998.
4. Neumann ES. Measurement of socket discomfort: part I: pressure sensation. J Prosthet Orthot. 2001;13:99-110..
5. Damasio A. Descartes' Error: Emotion, Reason, and the Human Brain. New York: Avon;1994:267.
6. Chaffin DB, Andersson GBJ, Martin BJ. Occupational Biomechanics, 3rd ed. New York: John Wiley; 1999:363.