-
New Forum Software Installed (Still a few issues but reopening the forum) More details.
New Screening Tool May Predict Chronic Pain Before It Starts, 2020, Brown and Lee
- Thread starter Milo
- Start date
The paper:
The FUTUREPAIN study: Validating a questionnaire to predict the probability of having chronic pain 7-10 years into the future
Abstract:
Objectives:
The FUTUREPAIN study develops a short general-purpose questionnaire, based on the biopsychosocial model, to predict the probability of developing or maintaining moderate-to-severe chronic pain 7–10 years into the future.
Methods:
This is a retrospective cohort study. Two-thirds of participants in the National Survey of Midlife Development in the United States were randomly assigned to a training cohort used to train a predictive machine learning model based on the least absolute shrinkage and selection operator (LASSO) algorithm, which produces a model with minimal covariates. Out-of-sample predictions from this model were then estimated using the remaining one-third testing cohort to determine the area under the receiver operating characteristic curve (AUROC). An optimal cut-point that maximized sensitivity and specificity was determined.
Results:
The LASSO model using 82 variables in the training cohort, yielded an 18-variable model with an out-of-sample AUROC of 0.85 (95% Confidence Interval (CI): 0.80, 0.91) in the testing cohort. The sum of sensitivity (0.88) and specificity (0.76) was maximized at a cut-point of 17 (95% CI: 15, 18) on a 0–100 scale where the AUROC was 0.82.
Discussion:
We developed a short general-purpose questionnaire that predicts the probability of an adult having moderate-to-severe chronic pain in 7-to-10 years. It has diagnostic ability greater than 80% and can be used regardless of whether a patient is currently experiencing chronic pain. Knowing which patients are likely to have moderate-to-severe chronic pain in the future allows clinicians to target preventive treatment
Open access: here
The FUTUREPAIN study: Validating a questionnaire to predict the probability of having chronic pain 7-10 years into the future
Abstract:
Objectives:
The FUTUREPAIN study develops a short general-purpose questionnaire, based on the biopsychosocial model, to predict the probability of developing or maintaining moderate-to-severe chronic pain 7–10 years into the future.
Methods:
This is a retrospective cohort study. Two-thirds of participants in the National Survey of Midlife Development in the United States were randomly assigned to a training cohort used to train a predictive machine learning model based on the least absolute shrinkage and selection operator (LASSO) algorithm, which produces a model with minimal covariates. Out-of-sample predictions from this model were then estimated using the remaining one-third testing cohort to determine the area under the receiver operating characteristic curve (AUROC). An optimal cut-point that maximized sensitivity and specificity was determined.
Results:
The LASSO model using 82 variables in the training cohort, yielded an 18-variable model with an out-of-sample AUROC of 0.85 (95% Confidence Interval (CI): 0.80, 0.91) in the testing cohort. The sum of sensitivity (0.88) and specificity (0.76) was maximized at a cut-point of 17 (95% CI: 15, 18) on a 0–100 scale where the AUROC was 0.82.
Discussion:
We developed a short general-purpose questionnaire that predicts the probability of an adult having moderate-to-severe chronic pain in 7-to-10 years. It has diagnostic ability greater than 80% and can be used regardless of whether a patient is currently experiencing chronic pain. Knowing which patients are likely to have moderate-to-severe chronic pain in the future allows clinicians to target preventive treatment
Open access: here
I think this had the potential to be a good study. I liked that they said that by identifying biological, social and psychological factors for a particular person, that would enable better targeted care to prevent future pain.
There are a number of problems though. Sorry that this goes on a bit and it's complicated.
There were 82 possible variables. And in those 82 variables, there were many ways to add or ignore parts of the components making up those variables.
There isn't any indication of how important each of the factors was to the model. There were 17 variables used in the model, making the survey instrument to predict future pain. But, we don't know what the most important 5 or 10 variables were. Why didn't they run a model with just a handful of parameters - or just the biological variables?
Biological variables - 8 of the 11 biological variables made it into the final model
Of the 8 possible clinical variables - 6 of them made it into the model (body mass index (with the maximum BMI being over 50); sleep quality; number of vehicle accidents causing injury; number of head injuries; number of joint injuries; number of surgeries. The only two not making the cut were 'do you have a low pain threshold? and 'number of broken bones' (which perhaps was too highly correlated with the other variables covering injuries and surgeries).
Another biological variable included in the model was 'do you have chronic pain now?', as was a question about current physical health.
The physical health question was a bit weird, illustrating what I mean about fiddling with the variables. It was a 5 point scale, ranging from 'Poor health'; 'Fair health', 'Good health', 'Very good health', to 'Excellent health'. 'Poor health', 'good health' and 'very good health' are all scored as 0. 'Fair health' contributes to the overall score that predicts future pain; 'Excellent health' reduces the overall score a bit. Ignoring poor health ratings when fair health ratings were used in the prediction seems to be cherry picking.
I'm pretty sure that there could have been more biological variables that would have have been useful for predicting future pain.
Social/demographic variables - there were 39 social variables to select from. Only 5 made it into the final model.
Employment status - being unemployed did not make future pain more likely and neither did being in the work force. Being permanently out of the work force did though. Probably there is some correlation between being permanently out of the work force and current health status.
There were questions on death in the family in the last 5 years, parental abuse, ever having lost a home to a fire or a flood and ever having suffered financial or property loss related to work. These questions were asked retrospectively and parental abuse covered a lot of ground include swearing and stomping out of the room.
I would have thought age would have been a significant predictor of future pain, given the age range at the end of the study was 35 to 83. 'Age' didn't make it into the model, but surely it affected the likelihood of deaths in the family (which included parents), and other losses over a lifetime.
This study presumably found no correlation between age, gender, education or income on future pain levels - none made it into the predictive model.
Psychological variables - there were 20 to choose from, with 4 making it into the model.
Notably, none of the big five personality traits made it into the model. So, it didn't matter how neurotic or disagreeable people were, that didn't predict future pain. This surely is a finding worth talking about. None of the big five personality traits were correlated with future pain levels! But it isn't discussed. Just in case you missed it - being a happy agreeable calm person wasn't found to lower the risk of pain in 7 to 10 years' time.
Also notably, a measure called 'somatic amplification' wasn't correlated with future pain either.
'General distress - depressive symptoms' wasn't correlated. 'General distress - anxious symptoms' wasn't correlated. None of four different measures of anger made it in. !!!!
So, what did?
The authors go to some effort to assure us that the fact that so many of the psychological variables didn't make it into the model doesn't mean they aren't correlated with future pain:
The way they combined the variables in the model seems quite odd. With the biological and social variables, the combinations of scores (mostly 1 or 0) and coefficients produce contributions to the total score of up to 6 points per variable. Even with the BMI, they multiply it by 0.22, to produce a figure in that range. For the 4 psychological variables though, the contributions to the total score are way higher.
To give you an example, Q9 is the Anxious arousal variable. There are 17 questions, 'How much of the time over the last week did you feel ... e.g. 'short of breath'. 'Not at all' scores 1. The lowest possible score for Q9 is 17. The coefficient is around 0.5, so if you never feel short of breath, you still add 8.5 to your score, swamping the contributions of the biological questions.
They could have converted the answers to a z value relating values to the population mean - so that the contributions to the scores of each question was better balanced.
There's more that could be said, but I'm having troubling writing. This study suggests that a whole range of personality and trauma factors that have been claimed to affect chronic illness and pain are not likely to affect future pain levels in people similar to this cohort. Much of the impact of the four psychological measures identified as being correlated here could be due to the overlap with biological variables (e.g. having trouble breathing could be a sign of ill health rather than anxiety).
There are a number of problems though. Sorry that this goes on a bit and it's complicated.
There were 82 possible variables. And in those 82 variables, there were many ways to add or ignore parts of the components making up those variables.
There isn't any indication of how important each of the factors was to the model. There were 17 variables used in the model, making the survey instrument to predict future pain. But, we don't know what the most important 5 or 10 variables were. Why didn't they run a model with just a handful of parameters - or just the biological variables?
Biological variables - 8 of the 11 biological variables made it into the final model
Of the 8 possible clinical variables - 6 of them made it into the model (body mass index (with the maximum BMI being over 50); sleep quality; number of vehicle accidents causing injury; number of head injuries; number of joint injuries; number of surgeries. The only two not making the cut were 'do you have a low pain threshold? and 'number of broken bones' (which perhaps was too highly correlated with the other variables covering injuries and surgeries).
Another biological variable included in the model was 'do you have chronic pain now?', as was a question about current physical health.
The physical health question was a bit weird, illustrating what I mean about fiddling with the variables. It was a 5 point scale, ranging from 'Poor health'; 'Fair health', 'Good health', 'Very good health', to 'Excellent health'. 'Poor health', 'good health' and 'very good health' are all scored as 0. 'Fair health' contributes to the overall score that predicts future pain; 'Excellent health' reduces the overall score a bit. Ignoring poor health ratings when fair health ratings were used in the prediction seems to be cherry picking.
I'm pretty sure that there could have been more biological variables that would have have been useful for predicting future pain.
Social/demographic variables - there were 39 social variables to select from. Only 5 made it into the final model.
Employment status - being unemployed did not make future pain more likely and neither did being in the work force. Being permanently out of the work force did though. Probably there is some correlation between being permanently out of the work force and current health status.
There were questions on death in the family in the last 5 years, parental abuse, ever having lost a home to a fire or a flood and ever having suffered financial or property loss related to work. These questions were asked retrospectively and parental abuse covered a lot of ground include swearing and stomping out of the room.
As mentioned before, they fiddled with the variables, which may have made variables that weren't correlated into variables that were.
I would have thought age would have been a significant predictor of future pain, given the age range at the end of the study was 35 to 83. 'Age' didn't make it into the model, but surely it affected the likelihood of deaths in the family (which included parents), and other losses over a lifetime.
This study presumably found no correlation between age, gender, education or income on future pain levels - none made it into the predictive model.
Psychological variables - there were 20 to choose from, with 4 making it into the model.
Notably, none of the big five personality traits made it into the model. So, it didn't matter how neurotic or disagreeable people were, that didn't predict future pain. This surely is a finding worth talking about. None of the big five personality traits were correlated with future pain levels! But it isn't discussed. Just in case you missed it - being a happy agreeable calm person wasn't found to lower the risk of pain in 7 to 10 years' time.
Also notably, a measure called 'somatic amplification' wasn't correlated with future pain either.
'General distress - depressive symptoms' wasn't correlated. 'General distress - anxious symptoms' wasn't correlated. None of four different measures of anger made it in. !!!!
So, what did?
- The Kessler K6 measure - They say it's a measure of non-specific psychological distress. It has six questions asking about how often people felt depressed, nervous, restless, hopeless, worthless and that everything was an effort. I imagine people currently with chronic pain might be more likely to feel more depressed or that everything was an effort.
- Anxious arousal - 17 questions about symptoms of anxiety, some of which could relate to physical causes e.g. trouble breathing, feeling faint.
- Loss of interest - 8 questions, covering things like feeling unattractive, feeling withdrawn from other people, feeling slowed down, thinking about death.
The authors go to some effort to assure us that the fact that so many of the psychological variables didn't make it into the model doesn't mean they aren't correlated with future pain:
- The last psychological measure is of religiosity and 'spiritual daily guidance' which was found to be protective against future pain. 'Problem solving coping' wasn't correlated though.
The way they combined the variables in the model seems quite odd. With the biological and social variables, the combinations of scores (mostly 1 or 0) and coefficients produce contributions to the total score of up to 6 points per variable. Even with the BMI, they multiply it by 0.22, to produce a figure in that range. For the 4 psychological variables though, the contributions to the total score are way higher.
To give you an example, Q9 is the Anxious arousal variable. There are 17 questions, 'How much of the time over the last week did you feel ... e.g. 'short of breath'. 'Not at all' scores 1. The lowest possible score for Q9 is 17. The coefficient is around 0.5, so if you never feel short of breath, you still add 8.5 to your score, swamping the contributions of the biological questions.
They could have converted the answers to a z value relating values to the population mean - so that the contributions to the scores of each question was better balanced.
There's more that could be said, but I'm having troubling writing. This study suggests that a whole range of personality and trauma factors that have been claimed to affect chronic illness and pain are not likely to affect future pain levels in people similar to this cohort. Much of the impact of the four psychological measures identified as being correlated here could be due to the overlap with biological variables (e.g. having trouble breathing could be a sign of ill health rather than anxiety).
Last edited:
Thanks Hutan, that's helpful. I couldn't face wading through the whole thing.
So it seems in sum, predictors of future pain are already existing physical caused of pain, and the few psychosocial 'causes' are likely the result of those existing pain and illnesses.
If I understand it correctly, the most important outcome, then, is that personality type and past psychosocial experiences don't predict future pain. Or have I oversimplified too far?
So it seems in sum, predictors of future pain are already existing physical caused of pain, and the few psychosocial 'causes' are likely the result of those existing pain and illnesses.
If I understand it correctly, the most important outcome, then, is that personality type and past psychosocial experiences don't predict future pain. Or have I oversimplified too far?
rvallee
Senior Member (Voting Rights)
It seems like the psychosocial aspects are completely superfluous here and basically past pain and accidents make it more likely that over time pain will become more painful. So basically low-level acute pain that does not resolve becomes low-level chronic pain then disabling chronic pain. So entirely as expected and no need to bring in woo. And yet, woo was brought.
Sorry but the religious question is clearly related to protestant culture in the US, which encourages to hide personal problems and work through hardship, meaning there are fewer reports simply because people report less. This is a known thing and a real problem because it makes preventative health care impossible. It's entirely superfluous and misleading to fall for such a random correlation.
The specificity here is basically zero, this is far too superficial. Retrospectively it may show tendencies but this has zero predictive ability on individual cases. Of course the beauty of the EBM approach is that it's only possible to test whether this works by using it in practice for something like a decade. Then when it is clearly shown to be useless at predicting anything it simply remains used in practice because, you know it: "it's been used for years and we like it". Escalation of commitment means committing to failure. Failure compounds, it brings other failure along with it. Past failure merges with present and future failure into a smorgasbord of disaster, all swept under the rug by simply not keeping track of anything relevant.
I will never laugh again at people who believed in stuff like the humors, the four elements and such. There are people today who believe in stuff every bit as dumb despite having everything in their grasp to know better. The number of deus ex machinas in medicine is way too damn high.
If they can stick with the relevant physical stuff this may be of use, or at least be a step forward. Clearly the only purpose of the BPS ideology is to make the BPS ideology relevant. It serves no other purpose.
Sorry but the religious question is clearly related to protestant culture in the US, which encourages to hide personal problems and work through hardship, meaning there are fewer reports simply because people report less. This is a known thing and a real problem because it makes preventative health care impossible. It's entirely superfluous and misleading to fall for such a random correlation.
The specificity here is basically zero, this is far too superficial. Retrospectively it may show tendencies but this has zero predictive ability on individual cases. Of course the beauty of the EBM approach is that it's only possible to test whether this works by using it in practice for something like a decade. Then when it is clearly shown to be useless at predicting anything it simply remains used in practice because, you know it: "it's been used for years and we like it". Escalation of commitment means committing to failure. Failure compounds, it brings other failure along with it. Past failure merges with present and future failure into a smorgasbord of disaster, all swept under the rug by simply not keeping track of anything relevant.
I will never laugh again at people who believed in stuff like the humors, the four elements and such. There are people today who believe in stuff every bit as dumb despite having everything in their grasp to know better. The number of deus ex machinas in medicine is way too damn high.
If they can stick with the relevant physical stuff this may be of use, or at least be a step forward. Clearly the only purpose of the BPS ideology is to make the BPS ideology relevant. It serves no other purpose.
I think you are spot on Trish. In fact, I made a mental note that this paper could be used to support arguments that personality type doesn't cause chronic illness.
Indeed. Though I was surprised that 'social' factors weren't more relevant - e.g. poverty leading to lack of access to the good medical care that stops acute issues becoming chronic, and a lifetime of manual work resulting in worn-out painful bodies. I expected more 'social' issues to go along with the 'bio' in affecting future pain. The fact that the sample was 93% white and 76% had some college education probably accounts for social issues not being more important predictors.
Last edited: