Identification of subgroups of patients with low back pain using Latent Class Analysis

Research output: Book/anthology/thesis/reportPh.D. thesisResearch

Abstract

Low back pain (LBP) is a major global health problem but the evidence base available to inform clinical decision making and to provide prognostic information to patients, is less than ideal. Therefore, there is a need for further knowledge about this largely non‐specific condition. Within this thesis, the approach taken to address this issue was by identifying LBP subgroups with the long‐term aim of being able to target treatment and provide more precise prognostic estimates. Data from 928 LBP patients, who participated in a cohort study in chiropractic practice, were used in this project.

The relationship between prognostic factors may vary from one subgroup to another, and most of the previous subgrouping studies, have not taken this into account. For example, hypothetically there may be a strong relationship in one subgroup between fear of movement and activity limitation, whereas this relationship may be weak or absent in another subgroup with different patient characteristics. Pattern recognition is a statistical method, which takes this into account by searching for relationships within the data, which in this project consisted of the patients’ reply to a comprehensive baseline questionnaire and the clinicians’ findings on a standardised examination of the low back. By using pattern recognition, subgroups of patients were identified within which their responses and scores are similar, and therefore the patients are more alike within the subgroups than across the subgroups.

Latent Class Analysis (LCA) can handle this kind of subgrouping and within this project it was used in a novel and explorative way. Subsequently, one aspect of the subgroupings external validity was investigated by testing their association with outcomes over twelve months using regression analyses. However, the optimal application of the LCA method in this context is unknown and therefore, two methodological considerations were addressed during the process. Firstly, when using existing questionnaire data, whether using each single item or the summary scores would provide better subgroup information. Secondly, whether using a two‐stage LCA approach (using health domains as an intermediate step) would increase the interpretability of the identified subgroups compared to a traditional single‐stage LCA approach (all items analysed simultaneously).

In this PhD project, the strategy of using each single item from the questionnaires was preferred, due to the more nuanced description available within the resulting subgroups. Therefore, the single‐item strategy was used in the subsequent single‐stage and two‐stage LCA, which identified seven and nine patient subgroups, respectively, with similar face validity and adequate statistical performance. Subsequent regression analyses showed a significant association between these subgroupings and the outcomes of LBP intensity and pain‐related disability, and their prognostic capacity was similar, albeit limited.

Although the subgroupings explained only a small additional variation in the outcomes, their prognostic capacity was as high or higher than two existing subgrouping tools (STarT Back Tool and Quebec Task Force Classification), and three baseline characteristics (LBP intensity, leg pain intensity and pain‐related disability). In contrast, the novel subgroupings had a lower prognostic capacity than a single question about the participants’ recovery beliefs, as well as lower than that of the domainspecific patient categorisations identified in the first stage of the two‐stage LCA.

Both of the LCA approaches identified distinct subgroups of LBP patients with some prognostic capacity. Therefore, further research could investigate whether any of these subgroup differences, either at the final subgroup level or at the domain‐specific level, have treatment implications or indicate differences in causal pathways.
Original languageEnglish
Place of PublicationPrint & Sign, SDU
PublisherSyddansk Universitet
Number of pages678
Publication statusPublished - 14. Oct 2016

Fingerprint

Low Back Pain
Regression Analysis
Chiropractic
Quebec
Advisory Committees
Reproducibility of Results
Leg
Cohort Studies
Health
Research

Cite this

@phdthesis{f6e03b784b4940cf8d878d80b7e89c50,
title = "Identification of subgroups of patients with low back pain using Latent Class Analysis",
abstract = "Low back pain (LBP) is a major global health problem but the evidence base available to inform clinical decision making and to provide prognostic information to patients, is less than ideal. Therefore, there is a need for further knowledge about this largely non‐specific condition. Within this thesis, the approach taken to address this issue was by identifying LBP subgroups with the long‐term aim of being able to target treatment and provide more precise prognostic estimates. Data from 928 LBP patients, who participated in a cohort study in chiropractic practice, were used in this project. The relationship between prognostic factors may vary from one subgroup to another, and most of the previous subgrouping studies, have not taken this into account. For example, hypothetically there may be a strong relationship in one subgroup between fear of movement and activity limitation, whereas this relationship may be weak or absent in another subgroup with different patient characteristics. Pattern recognition is a statistical method, which takes this into account by searching for relationships within the data, which in this project consisted of the patients’ reply to a comprehensive baseline questionnaire and the clinicians’ findings on a standardised examination of the low back. By using pattern recognition, subgroups of patients were identified within which their responses and scores are similar, and therefore the patients are more alike within the subgroups than across the subgroups.Latent Class Analysis (LCA) can handle this kind of subgrouping and within this project it was used in a novel and explorative way. Subsequently, one aspect of the subgroupings external validity was investigated by testing their association with outcomes over twelve months using regression analyses. However, the optimal application of the LCA method in this context is unknown and therefore, two methodological considerations were addressed during the process. Firstly, when using existing questionnaire data, whether using each single item or the summary scores would provide better subgroup information. Secondly, whether using a two‐stage LCA approach (using health domains as an intermediate step) would increase the interpretability of the identified subgroups compared to a traditional single‐stage LCA approach (all items analysed simultaneously).In this PhD project, the strategy of using each single item from the questionnaires was preferred, due to the more nuanced description available within the resulting subgroups. Therefore, the single‐item strategy was used in the subsequent single‐stage and two‐stage LCA, which identified seven and nine patient subgroups, respectively, with similar face validity and adequate statistical performance. Subsequent regression analyses showed a significant association between these subgroupings and the outcomes of LBP intensity and pain‐related disability, and their prognostic capacity was similar, albeit limited.Although the subgroupings explained only a small additional variation in the outcomes, their prognostic capacity was as high or higher than two existing subgrouping tools (STarT Back Tool and Quebec Task Force Classification), and three baseline characteristics (LBP intensity, leg pain intensity and pain‐related disability). In contrast, the novel subgroupings had a lower prognostic capacity than a single question about the participants’ recovery beliefs, as well as lower than that of the domainspecific patient categorisations identified in the first stage of the two‐stage LCA.Both of the LCA approaches identified distinct subgroups of LBP patients with some prognostic capacity. Therefore, further research could investigate whether any of these subgroup differences, either at the final subgroup level or at the domain‐specific level, have treatment implications or indicate differences in causal pathways.",
author = "Nielsen, {Anne M{\o}lgaard}",
year = "2016",
month = "10",
day = "14",
language = "English",
publisher = "Syddansk Universitet",
address = "Denmark",

}

Identification of subgroups of patients with low back pain using Latent Class Analysis. / Nielsen, Anne Mølgaard.

Print & Sign, SDU : Syddansk Universitet, 2016. 678 p.

Research output: Book/anthology/thesis/reportPh.D. thesisResearch

TY - BOOK

T1 - Identification of subgroups of patients with low back pain using Latent Class Analysis

AU - Nielsen, Anne Mølgaard

PY - 2016/10/14

Y1 - 2016/10/14

N2 - Low back pain (LBP) is a major global health problem but the evidence base available to inform clinical decision making and to provide prognostic information to patients, is less than ideal. Therefore, there is a need for further knowledge about this largely non‐specific condition. Within this thesis, the approach taken to address this issue was by identifying LBP subgroups with the long‐term aim of being able to target treatment and provide more precise prognostic estimates. Data from 928 LBP patients, who participated in a cohort study in chiropractic practice, were used in this project. The relationship between prognostic factors may vary from one subgroup to another, and most of the previous subgrouping studies, have not taken this into account. For example, hypothetically there may be a strong relationship in one subgroup between fear of movement and activity limitation, whereas this relationship may be weak or absent in another subgroup with different patient characteristics. Pattern recognition is a statistical method, which takes this into account by searching for relationships within the data, which in this project consisted of the patients’ reply to a comprehensive baseline questionnaire and the clinicians’ findings on a standardised examination of the low back. By using pattern recognition, subgroups of patients were identified within which their responses and scores are similar, and therefore the patients are more alike within the subgroups than across the subgroups.Latent Class Analysis (LCA) can handle this kind of subgrouping and within this project it was used in a novel and explorative way. Subsequently, one aspect of the subgroupings external validity was investigated by testing their association with outcomes over twelve months using regression analyses. However, the optimal application of the LCA method in this context is unknown and therefore, two methodological considerations were addressed during the process. Firstly, when using existing questionnaire data, whether using each single item or the summary scores would provide better subgroup information. Secondly, whether using a two‐stage LCA approach (using health domains as an intermediate step) would increase the interpretability of the identified subgroups compared to a traditional single‐stage LCA approach (all items analysed simultaneously).In this PhD project, the strategy of using each single item from the questionnaires was preferred, due to the more nuanced description available within the resulting subgroups. Therefore, the single‐item strategy was used in the subsequent single‐stage and two‐stage LCA, which identified seven and nine patient subgroups, respectively, with similar face validity and adequate statistical performance. Subsequent regression analyses showed a significant association between these subgroupings and the outcomes of LBP intensity and pain‐related disability, and their prognostic capacity was similar, albeit limited.Although the subgroupings explained only a small additional variation in the outcomes, their prognostic capacity was as high or higher than two existing subgrouping tools (STarT Back Tool and Quebec Task Force Classification), and three baseline characteristics (LBP intensity, leg pain intensity and pain‐related disability). In contrast, the novel subgroupings had a lower prognostic capacity than a single question about the participants’ recovery beliefs, as well as lower than that of the domainspecific patient categorisations identified in the first stage of the two‐stage LCA.Both of the LCA approaches identified distinct subgroups of LBP patients with some prognostic capacity. Therefore, further research could investigate whether any of these subgroup differences, either at the final subgroup level or at the domain‐specific level, have treatment implications or indicate differences in causal pathways.

AB - Low back pain (LBP) is a major global health problem but the evidence base available to inform clinical decision making and to provide prognostic information to patients, is less than ideal. Therefore, there is a need for further knowledge about this largely non‐specific condition. Within this thesis, the approach taken to address this issue was by identifying LBP subgroups with the long‐term aim of being able to target treatment and provide more precise prognostic estimates. Data from 928 LBP patients, who participated in a cohort study in chiropractic practice, were used in this project. The relationship between prognostic factors may vary from one subgroup to another, and most of the previous subgrouping studies, have not taken this into account. For example, hypothetically there may be a strong relationship in one subgroup between fear of movement and activity limitation, whereas this relationship may be weak or absent in another subgroup with different patient characteristics. Pattern recognition is a statistical method, which takes this into account by searching for relationships within the data, which in this project consisted of the patients’ reply to a comprehensive baseline questionnaire and the clinicians’ findings on a standardised examination of the low back. By using pattern recognition, subgroups of patients were identified within which their responses and scores are similar, and therefore the patients are more alike within the subgroups than across the subgroups.Latent Class Analysis (LCA) can handle this kind of subgrouping and within this project it was used in a novel and explorative way. Subsequently, one aspect of the subgroupings external validity was investigated by testing their association with outcomes over twelve months using regression analyses. However, the optimal application of the LCA method in this context is unknown and therefore, two methodological considerations were addressed during the process. Firstly, when using existing questionnaire data, whether using each single item or the summary scores would provide better subgroup information. Secondly, whether using a two‐stage LCA approach (using health domains as an intermediate step) would increase the interpretability of the identified subgroups compared to a traditional single‐stage LCA approach (all items analysed simultaneously).In this PhD project, the strategy of using each single item from the questionnaires was preferred, due to the more nuanced description available within the resulting subgroups. Therefore, the single‐item strategy was used in the subsequent single‐stage and two‐stage LCA, which identified seven and nine patient subgroups, respectively, with similar face validity and adequate statistical performance. Subsequent regression analyses showed a significant association between these subgroupings and the outcomes of LBP intensity and pain‐related disability, and their prognostic capacity was similar, albeit limited.Although the subgroupings explained only a small additional variation in the outcomes, their prognostic capacity was as high or higher than two existing subgrouping tools (STarT Back Tool and Quebec Task Force Classification), and three baseline characteristics (LBP intensity, leg pain intensity and pain‐related disability). In contrast, the novel subgroupings had a lower prognostic capacity than a single question about the participants’ recovery beliefs, as well as lower than that of the domainspecific patient categorisations identified in the first stage of the two‐stage LCA.Both of the LCA approaches identified distinct subgroups of LBP patients with some prognostic capacity. Therefore, further research could investigate whether any of these subgroup differences, either at the final subgroup level or at the domain‐specific level, have treatment implications or indicate differences in causal pathways.

M3 - Ph.D. thesis

BT - Identification of subgroups of patients with low back pain using Latent Class Analysis

PB - Syddansk Universitet

CY - Print & Sign, SDU

ER -