Limits...
Overcoming pain thresholds with multilevel models-an example using quantitative sensory testing (QST) data.

Hirschfeld G, Blankenburg MR, Süß M, Zernikow B - PeerJ (2015)

Bottom Line: Application of these methods led to intriguing findings, such as the presence lower pain-thresholds in healthy children compared to healthy adolescents.We describe how multilevel models can be used to estimate these parameters and to overcome central critiques of these methods.Overall, we hope that the wider use of multilevel modeling to describe somatosensory functioning may advance neurology research and practice.

View Article: PubMed Central - HTML - PubMed

Affiliation: Faculty of Business Management and Social Sciences, University of Applied Sciences Osnabrück , Osnabrück , Germany.

ABSTRACT
The assessment of somatosensory function is a cornerstone of research and clinical practice in neurology. Recent initiatives have developed novel protocols for quantitative sensory testing (QST). Application of these methods led to intriguing findings, such as the presence lower pain-thresholds in healthy children compared to healthy adolescents. In this article, we (re-) introduce the basic concepts of signal detection theory (SDT) as a method to investigate such differences in somatosensory function in detail. SDT describes participants' responses according to two parameters, sensitivity and response-bias. Sensitivity refers to individuals' ability to discriminate between painful and non-painful stimulations. Response-bias refers to individuals' criterion for giving a "painful" response. We describe how multilevel models can be used to estimate these parameters and to overcome central critiques of these methods. To provide an example we apply these methods to data from the mechanical pain sensitivity test of the QST protocol. The results show that adolescents are more sensitive to mechanical pain and contradict the idea that younger children simply use more lenient criteria to report pain. Overall, we hope that the wider use of multilevel modeling to describe somatosensory functioning may advance neurology research and practice.

No MeSH data available.


Related in: MedlinePlus

Responses and fitted model for one subject.Participants rated stimuli at different intensities. Note, points represent the average % of responses painful at each stimulus intensity, the blue line indicates the fit to the individual data.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4636408&req=5

fig-2: Responses and fitted model for one subject.Participants rated stimuli at different intensities. Note, points represent the average % of responses painful at each stimulus intensity, the blue line indicates the fit to the individual data.

Mentions: An alternative to describing participants’ responses in terms of thresholds is to use signal detection theory to model their responses within individual trials (Green & Swets, 1966; Macmillan & Creelman, 2004). However, to use this method, data need to be collected using the method of constant stimuli (gathering individual ratings on a random series of stimuli, including empty trials without stimulation). According to signal detection theory, participants’ responses can be modeled by two parameters, sensitivity and response bias. Sensitivity refers to the ability to accurately discriminate between the presence and absence of a target stimulus. Response bias refers to participants’ criteria for reporting the presence or absence of the stimulus. Differences in these parameters become apparent when stimulus-response-functions are plotted for a number of participants (see Fig. 2 below). Participants with high sensitivity show a steep increase in the percent of painful responses with increasing stimulus intensities. Participants with high criteria for pain show these increases at higher stimulus intensities. The most widely used metrics for sensitivity and response bias are d′ and c, respectively. These two parameters are readily calculated from a 2 × 2 table consisting of the counts for hits (participants correctly report presence of stimuli), misses (participants report absence of stimuli even though they are present), false-alarms (participants report presence of stimuli even though they are absent), and correct rejections (participants correctly report absence of stimuli) (Green & Swets, 1966). From these the hit rate (hits/(hits + misses)) and false-alarm rate (false-alarms/(false alarms + correct rejections)) are computed. The measure d′ is defined as the z-score corresponding to the hit rate minus the z-score corresponding to the false-alarm rate. If participants are very good at distinguishing the presence or absence of stimuli, they have many hits and correct rejections and few false-alarms and misses resulting in high values for d′. If participants are bad at distinguishing the presence or absence of stimuli and perform at chance level, d′ will be zero. The measure c is defined as −.5 (-score corresponding to the hit rate plus the z-score corresponding to the false-alarm rate). If participants show no preference for either response, c is zero. If participants have a preference towards a specific response this will be either positive or negative. More information on the calculation and interpretation of SDT measures is given by Stanislaw & Todorov (1999).


Overcoming pain thresholds with multilevel models-an example using quantitative sensory testing (QST) data.

Hirschfeld G, Blankenburg MR, Süß M, Zernikow B - PeerJ (2015)

Responses and fitted model for one subject.Participants rated stimuli at different intensities. Note, points represent the average % of responses painful at each stimulus intensity, the blue line indicates the fit to the individual data.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4636408&req=5

fig-2: Responses and fitted model for one subject.Participants rated stimuli at different intensities. Note, points represent the average % of responses painful at each stimulus intensity, the blue line indicates the fit to the individual data.
Mentions: An alternative to describing participants’ responses in terms of thresholds is to use signal detection theory to model their responses within individual trials (Green & Swets, 1966; Macmillan & Creelman, 2004). However, to use this method, data need to be collected using the method of constant stimuli (gathering individual ratings on a random series of stimuli, including empty trials without stimulation). According to signal detection theory, participants’ responses can be modeled by two parameters, sensitivity and response bias. Sensitivity refers to the ability to accurately discriminate between the presence and absence of a target stimulus. Response bias refers to participants’ criteria for reporting the presence or absence of the stimulus. Differences in these parameters become apparent when stimulus-response-functions are plotted for a number of participants (see Fig. 2 below). Participants with high sensitivity show a steep increase in the percent of painful responses with increasing stimulus intensities. Participants with high criteria for pain show these increases at higher stimulus intensities. The most widely used metrics for sensitivity and response bias are d′ and c, respectively. These two parameters are readily calculated from a 2 × 2 table consisting of the counts for hits (participants correctly report presence of stimuli), misses (participants report absence of stimuli even though they are present), false-alarms (participants report presence of stimuli even though they are absent), and correct rejections (participants correctly report absence of stimuli) (Green & Swets, 1966). From these the hit rate (hits/(hits + misses)) and false-alarm rate (false-alarms/(false alarms + correct rejections)) are computed. The measure d′ is defined as the z-score corresponding to the hit rate minus the z-score corresponding to the false-alarm rate. If participants are very good at distinguishing the presence or absence of stimuli, they have many hits and correct rejections and few false-alarms and misses resulting in high values for d′. If participants are bad at distinguishing the presence or absence of stimuli and perform at chance level, d′ will be zero. The measure c is defined as −.5 (-score corresponding to the hit rate plus the z-score corresponding to the false-alarm rate). If participants show no preference for either response, c is zero. If participants have a preference towards a specific response this will be either positive or negative. More information on the calculation and interpretation of SDT measures is given by Stanislaw & Todorov (1999).

Bottom Line: Application of these methods led to intriguing findings, such as the presence lower pain-thresholds in healthy children compared to healthy adolescents.We describe how multilevel models can be used to estimate these parameters and to overcome central critiques of these methods.Overall, we hope that the wider use of multilevel modeling to describe somatosensory functioning may advance neurology research and practice.

View Article: PubMed Central - HTML - PubMed

Affiliation: Faculty of Business Management and Social Sciences, University of Applied Sciences Osnabrück , Osnabrück , Germany.

ABSTRACT
The assessment of somatosensory function is a cornerstone of research and clinical practice in neurology. Recent initiatives have developed novel protocols for quantitative sensory testing (QST). Application of these methods led to intriguing findings, such as the presence lower pain-thresholds in healthy children compared to healthy adolescents. In this article, we (re-) introduce the basic concepts of signal detection theory (SDT) as a method to investigate such differences in somatosensory function in detail. SDT describes participants' responses according to two parameters, sensitivity and response-bias. Sensitivity refers to individuals' ability to discriminate between painful and non-painful stimulations. Response-bias refers to individuals' criterion for giving a "painful" response. We describe how multilevel models can be used to estimate these parameters and to overcome central critiques of these methods. To provide an example we apply these methods to data from the mechanical pain sensitivity test of the QST protocol. The results show that adolescents are more sensitive to mechanical pain and contradict the idea that younger children simply use more lenient criteria to report pain. Overall, we hope that the wider use of multilevel modeling to describe somatosensory functioning may advance neurology research and practice.

No MeSH data available.


Related in: MedlinePlus