Professional Documents
Culture Documents
Reductionist Research
Reductionist Research
net/publication/333143836
CITATIONS READS
0 23
4 authors, including:
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Trunk motor control performance before and after spinal manipulation treatment View project
Supporting Innovation in Team Science through Online Collaborative Modeling View project
All content following this page was uploaded by Pramod K Pathak on 22 May 2019.
L
ow back pain (LBP) is a multifactorial problem associated with “factors contributing to LBP”). This ap-
many biological, psychological, and social factors.8,21,28,30 In most proach is based on the rationale that
cases, the exact causes underlying LBP are unknown; hence, the when more is known about the etiology of
LBP, the treatment can be more specific
term nonspecific LBP is often used. This nonspecificity makes
in addressing the factors contributing to
selecting the appropriate treatment challenging for clinicians. Therefore, LBP and result in better outcomes. Subse-
much of the current research efforts are directed toward identifying quently, randomized clinical trials (RCTs)
specific causal factors underlying the that determine the nature of an individ- are conducted to evaluate whether such
clinical presentation of LBP or toward ual’s LBP) to formulate the appropriate matching between factors contributing
subclassifying patients with specific intervention strategies addressing these to LBP and treatment leads to improved
characteristics (a collection of factors specific factors (hereafter referred to as outcomes compared to other treatments,
standard care, or sham treatment.
The above-outlined strategy in LBP
UUBACKGROUND: Traditionally, low back pain tor versus a multimodal treatment that eliminates
(LBP) is studied using a reductionist approach, a number of the randomly selected factors. research is termed a reductionist ap-
proach in the parlance of systems sci-
in which the factors contributing to the clinical
UURESULTS: With an increasing number of factors,
presentation of LBP are studied in isolation to
the probability of subclassifying an individual to
ence.1 In the reductionist approach,
identify the primary pathology or condition linked the system is broken down into smaller
a subgroup based on a single factor tends toward
to LBP. We argue that reductionism may not be
zero. A multimodal treatment arbitrarily address- parts to isolate and study them compre-
suitable for studying LBP, considering the complex,
ing any 2 or more factors was more effective hensively. The reductionist approach is
multifactorial nature of this condition.
than diagnosing and treating a single factor that
UUOBJECTIVES: To quantify the likelihood of
well suited for containable diseases, such
maximally contributed to LBP.
as local infection. However, reduction-
UUCONCLUSION: Results suggest that reduc-
successfully subclassifying patients with LBP and
effectively targeting treatment based on a single ism is less helpful when the problem is
dominant factor contributing to LBP. tionism is not appropriate for subclassifying
multifactorial and where interactions
patients with LBP or for targeting treatment.
UUMETHODS: Both analytical and numerical The use of reductionist approaches may explain between biological subsystems exist. 1
simulations (Monte Carlo) of 1 million patients with some of the challenges when creating an LBP These features make the behavior of
LBP were performed. Several factors contributing classification system and designing effective a complex system difficult to predict,
to LBP were randomly assigned to each individual.
treatment interventions. J Orthop Sports Phys even when the behavior of its parts is
The following outcome measures were computed,
Ther 2019;49(6):477-481. Epub 15 May 2019.
as a function of the number of factors: the percent- well defined.1 For example, studying
doi:10.2519/jospt.2019.8791
age of individuals who could be subclassified by motor control in patients with LBP is
identifying a single factor exceeding a certain UUKEY WORDS: classification, Monte Carlo a reductionist approach that evaluates
threshold, and the average reduction in LBP when simulation, randomized clinical trials, risk factors,
the pathomechanics of neuromuscular
treatment eliminates the largest contributing fac- subgrouping
control in isolation from other biologi-
Center for Orthopedic Research, Michigan State University, Lansing, MI. 2Department of Osteopathic Surgical Specialties, Michigan State University, East Lansing, MI.
1
3
Department of Statistics and Probability, College of Natural Science, Michigan State University, East Lansing, MI. 4Sumaq Life LLC, East Lansing, MI. Drs Cholewicki, Reeves,
and Popovich, Jr. were partially supported by National Center for Complementary and Integrative Health grant U19AT006057-01A1 from the National Institutes of Health. This
paper’s contents are solely the responsibility of the authors and do not necessarily represent the official views of the National Center for Complementary and Integrative Health.
Dr Reeves is the founder and president of Sumaq Life LLC. The authors certify that they have no affiliations with or financial involvement in any organization or entity with a
direct financial interest in the subject matter or materials discussed in the article. Address correspondence to Dr Jacek Cholewicki, Center for Orthopedic Research, Michigan
State University, McLaren Orthopedic Hospital, 2727 South Pennsylvania Avenue, Lansing, MI 48910. E-mail: cholewic@msu.edu t Copyright ©2019 Journal of Orthopaedic &
Sports Physical Therapy®
W
correct those pathomechanics.27 shows low to moderate effects and prac- e performed both analytical
Reductionism is not inherently wrong, tically no differences between various in- and numerical simulations (Mon-
as it allows for the identification of parts terventions.3 More importantly, to date, te Carlo) of a large population (n
of the system (eg, factors associated with “no classification system is supported by = 1 million) with LBP. Factors contrib-
LBP) and has been useful for establish- sufficient evidence to recommend im- uting to LBP for each individual were
ing factors associated with patient pre- plementation into clinical practice.”2,7,13 uniformly distributed random variates
sentations (phenotypes), an important Even a triage based on various clinical (U1, U2, U3, ... Uk[AQ: X as in Fig 1?])
part of patient care. The problem lies in prediction rules has not led to better between 0 and 1 (FIGURE 1). For each in-
the assumption that information about outcomes.13 One possible explanation dividual, each factor Ui was normalized
individual parts is sufficient to explain for the lack of success in documenting by dividing it by the sum of k factors to
the behavior of the entire system. In the large positive treatment outcomes could create a fraction contribution to LBP; for
example of studying motor control using be the reductionist approach, typically example, the total pain/disability effect of
a reductionist approach, the assumption applied in LBP research, whereby uni- 1 is: (X1 + X2 + … + Xk) = 1. For example,
is that other biological, psychological, modal intervention strategies targeting for 3 factors (k = 3), a person with LBP
and social factors have minimal or no the dominant factor believed to be con- may have normalized factors such as X1
influence on motor control. Perhaps in tributing to LBP are compared and stud- = 0.3, X2 = 0.1, and X3 = 0.6. This means
some patients this may be the case, but ied in RCTs. While this approach has its that factor X1 contributes 30%, factor X2
the evidence suggests that motor control place in research, considering the ex- contributes 10%, and factor X3 contrib-
interventions are not superior to other treme biological complexity of the spine utes 60% to the overall presentation of
interventions in the management of pa- system, the multifactorial nature of LBP, LBP, totaling 100%. To test the 2 hypoth-
tients with LBP,27 which raises questions and interactions among these factors,21,30 eses, we calculated (1) the percentage of
about other factors and interactions an approach that addresses these issues individuals who could be subclassified
involved. simultaneously is needed to advance by identifying a single normalized fac-
In contrast to reductionism, a systems LBP research and the development of tor (Xi) exceeding a certain threshold θ
approach takes the entire system into more effective intervention strategies. (where θ = 0.2, 0.3, 0.4), and (2) the av-
consideration when describing its be- The goal of this study was to high- erage reduction in pain/disability when
havior and identifying interdependence light the challenges of studying a the largest factor contributing to LBP is
between its subsystems.1 Attempts to- complex condition using reductionist identified and eliminated with the target-
ward such an approach have been made approaches. Specifically, using analytical ed unimodal treatment, versus a number
with conceptual, structural equation, or and numerical simulations, we quanti- of treatments (multimodal treatment)
collaborative modeling to account for a fied the likelihood of correctly identify- eliminating a number of randomly se-
number of factors contributing to LBP ing the dominant factor contributing to lected factors.
simultaneously.5,6,9,18,25 Yet, research in LBP and of effectively treating LBP by
LBP lags substantially behind systems modifying such a dominant factor. The
biology, which rapidly progressed in re- following 2 hypotheses were tested: (1) LBP
cent years with its effective application when dealing with a large number of fac-
of systems science.4,14 There is a critical tors contributing to LBP, it is not pos-
lack of knowledge regarding the number sible to identify subgroups effectively
of factors and their interactions needed based on the dominant factor; and (2)
to adequately represent LBP, which in on a population scale, providing a num- X1 X2 Xi Xk
turn, limits the ability to target them ber of treatments targeting any 2 or
through treatment modalities. As spine more factors is more effective than iden-
FIGURE 1. A schematic of the multifactorial,
research evolves, the trend points to- tifying and treating a single factor that uniformly distributed model of LBP used in this study.
ward more complexity, with more sub- maximally contributes to LBP. If these All factors contributing to the clinical presentation
systems and their interactions requiring hypotheses are true, perhaps a different of LBP were independent. Note that the sum of all
consideration.6,11 research method, based on a systems ap- factors (Xi) contributing to LBP is equal to 1 in every
case simulated. Abbreviation: LBP, low back pain.
There have been more than 1000 proach,1 could lead to the development
T T
he maximum differences be- he results from our analytical of pure subgroups rare.16
tween any analytically derived val- and numerical simulations of a Based on the number of existing
ues and the 2 simulation results multifactorial presentation of LBP baseline predictors and the variance
were 5.28 × 10–4 and 4.75 × 10–4. These are consistent with the data reported in in outcomes they explain, Mistry et al19
small differences indicate excellent agree- the literature. With respect to the first concluded that it is unlikely we can iden-
ment between the 2 methods, validating hypothesis, our results show that with tify a single strong moderator of LBP
the analytical approach. an increasing number of factors con- treatment effects. None of the RCTs they
With an increasing number of fac- tributing to LBP, there is a diminishing reviewed were powered sufficiently to
tors, the probability of a single factor likelihood of classifying an individual identify differential subgroup effects, and
exceeding a certain threshold (Xi>θ) to a subgroup of patients based on the appropriately powered studies would be
tends toward zero (FIGURE 2). In our dominant factor. This could explain why practically unrealistic.19 To circumvent
model, this result represents the dimin- attempts to identify subgroups of patients this problem, Patel et al23 pooled data
ishing likelihood of classifying an indi- who would respond more favorably to a from 19 back pain trials that provided a
vidual to a subgroup of patients with data set of 9328 patients. Yet they, too,
LBP based on a single factor reaching 100 did not find any subgroups that would
some set threshold of contribution to benefit from specific treatment, and,
Reduction in Pain, %
80
the overall LBP (FIGURE 2). Even with a more importantly, they calculated that
low threshold of θ = 0.2 (accounting for 60 such an approach to identifying patients
20% of LBP symptoms), less than 1% of 40 would not be cost-effective.
the LBP population can be subclassified Our simulations are consistent with
when the number of factors exceeds 11. 20 such findings. With only 12 factors con-
0 tributing to LBP, only 0.5% of the LBP
0 3 6 9 12 15 18 population could be subclassified based
100
Factors Contributing to LBP, n on a single factor and treated to achieve
80
Maximum-factor Tx 1-factor Tx a minimal clinically important difference
Individuals, %
R
such a comparison. Alternatively, in the esearch to identify the factors, likely be more effective than unimodal
above example, the psychological factors or group of factors, that contribute treatment.
targeted by cognitive behavioral therapy to LBP and to understand the ef- CAUTION: The main assumptions influ-
might not have been important factors ficacy of individual treatment interven- encing the specific numerical results
contributing to LBP in these patients. tions is necessary but not sufficient to were that factors contributing to LBP
Several assumptions determine the address the LBP problem effectively. As were uniformly distributed and that
behavior of this model simulation. The demonstrated by our unstructured mul- there were no interactions among them.
assumption having probably the biggest tifactorial model of LBP, simply identi- While these assumptions affect the com-
effect on the results was that various fac- fying components within the model and plexity of the modeled LBP problem,
tors contributing to LBP are uniformly not the structure of the model (ie, the in- the simulation trends will likely hold for
distributed across the population with teractions between these components) is more complex models.
LBP. That is, all factors have the same not likely to lead to robust classification
probability of being present in each in- or better treatment effects. ACKNOWLEDGMENTS: The forum on which this
dividual, and there is no factor occurring To advance LBP research, more so- body of research was based, “State-of-the-Art
more frequently in the LBP population. phisticated modeling methods that con- in Motor Control and Low Back Pain: Inter-
If some factors were occurring more sider the structure of the system being national Clinical and Research Expert Fo-
frequently, it would have been easier to studied9,18 and possibly the dynamics of rum,” was supported by the National Health
identify a cluster of patients with these the system1 (LBP symptoms and treat- and Medical Research Council of Australia,
factors. We submit, however, that in real- ment effects are not static and change in collaboration with the North American
ity the distribution of factors contribut- with time) are needed. Future research Spine Society. The forum was chaired by Dr
ing to LBP might be closer to uniform, should involve a paradigm shift toward Paul Hodges.
@ MORE INFORMATION
ate takes values between 0 and 1, all such values 21. O ’Sullivan P, Caneiro JP, O’Keeffe M, O’Sullivan K.
being equally probable. Biometrika. 1927;19:240- Unraveling the complexity of low back pain. J Or-
245. https://doi.org/10.1093/biomet/19.3-4.240 thop Sports Phys Ther. 2016;46:932-937. https:// WWW.JOSPT.ORG
HYPOTHESIS 1
To estimate how many people can be subclassified based on identifying a single factor exceeding a certain threshold θ, we needed to calculate the
probability (P) of a factor X1>θ in the population with LBP. The following derivation is a consequence of the sampling distribution of a large number of
uniform variates on the unit interval (0, 1).
Let U1, U2, ... be independent, random variables uniformly distributed on the unit interval (0, 1). Let θ be a given number: 0<θ<1. Let Sk+1 = U1 + ... + Uk+1,
Xi = Ui/Sk+1, 1≤i≤k + 1.
(1 − )
= {(1 − ) 1 > } = < 1
(1 − ) (1 − ) (1 − )
( 1 > )= < 1 = < = < → 0 as → ∞
where u has uniform distribution on (0,1). For large k, the distribution of Sk is approximately normal, with mean k/2 and standard deviation √k/12
(Irwin-Hall distribution).10 Therefore,
1−
( 1 > )≈ Φ , ,
2 12
in which Φ(z, μ, σ) denotes the cumulative distribution function of the normal variate, with mean μ and standard deviation σ.
HYPOTHESIS 2
To address hypothesis 2, we must estimate the expected value of the sum of m factors E(X1 + X2 + ... + Xm) and the expected value of the maximum fac-
tor E(Xmax). Based on the same Irwin-Hall distribution,10
+1
( +1 ) =
2
and
1 2 +1 1 +1 1
= = ⋯ = = =
+1 +1 +1 +1 +1 +1
Therefore,
1 + ⋯+ 1
( 1+ ⋯ + ) = = =
+1 +1 +1
The expected value of the maximum factor is a ratio of 2 random variables, and to the first-order approximation is
max 1 (max ) 2
( ) = ≈ = (max ) =
+1 ( +1 ) +1
2 2( + 1) ( +1) ( +1)
2
= ( + 1) = = 2 = → 0 as → ∞
+1 +1 +2