You are on page 1of 8

1081

ORIGINAL ARTICLE

The Effects of Splinting on Outcomes for Epicondylitis


V. Jane Derebery, MD, Jenny N. Devenport, PhD, Geneva M. Giang, MBA, W. Tom Fogarty, MD
ABSTRACT. Derebery VJ, Devenport JN, Giang GM, in particular, its work-relatedness.4-13 A review of epidemio-
Fogarty WT. The effects of splinting on outcomes for epicon- logic studies of workplace factors by the US National Institute
dylitis. Arch Phys Med Rehabil 2005;86:1081-8. of Occupational Safety and Health (NIOSH) determined that
there was insufficient evidence for (1) an association between
Objective: To evaluate the effects of splinting on outcomes repetitive work and elbow musculoskeletal disorders and (2) an
for injured workers with epicondylitis. association between postural factors and epicondylitis.14 Both
Design: Retrospective cohort study using propensity score findings are contrary to common assumptions of both workers
methodology to statistically control for all observed pretreat- and physicians, which include, for instance, belief in a causal
ment differences between patients with and without splints. link between keyboarding and epicondylitis.15 The NIOSH
Setting: Nationwide network of 253 occupational medicine review concluded that there was evidence for an association
clinics. between forceful work and epicondylitis and strong evidence
Participants: All injured workers (N⫽4614) receiving pri- for a relationship between exposure to a combination of risk
mary care for lateral or medical epicondylitis (International factors (force, repetition, posture) and epicondylitis.14
Classification of Diseases, 9th Revision, codes 726.31 or Most patients with epicondylitis are managed by primary
726.32). care physicians, and common treatments include rest, ice, non-
Interventions: Not applicable. steroidal anti-inflammatory drugs (NSAIDs), corticosteroid
Main Outcome Measures: Physician-prescribed rates of injections, range of motion exercises, stretching and strength-
duty restrictions and lost time, treatment duration, specialist ening exercises, counterforce bracing and splinting, ionto-
referrals, and medical and physical therapy (PT) visits and phoresis, acupuncture, and ergonomic adjustments or training.
charges. Less commonly, refractory cases may be treated surgically.16-18
Results: Overall, patients with splints had higher rates of The long-term prognosis for epicondylitis is good, with 80% of
limited duty (P⬍.001), more medical visits and charges patients recovering within a year.19,20 However, there is no
(P⬍.001), higher total charges (medical and PT, P⬍.001), and strong consistent evidence to support any one treatment as
longer treatment durations (P⬍.01) than patients without being more efficacious in the long term than any other, includ-
splints. Evaluating differences for patients who did and did not ing “wait and see.”12,18,20-23 Recent research18,20 suggests that,
receive PT, significant differences remained for rates of limited in the short term (4 – 6wk), corticosteroid injections may out-
duty (P⬍.05), medical visits (P⬍.01), and medical charges perform physical therapy (PT), NSAIDs, and wait-and-see
(P⬍.01). treatments, but with longer follow-up, most patients improve
Conclusions: Splinting patients with epicondylitis may not regardless of treatment.
optimize outcomes, including rates of limited duty, treatment Study results for splinting, a commonly prescribed treat-
duration, and medical costs. ment, are also equivocal, with some studies yielding findings
Key Words: Epicondylitis; Occupational medicine; Reha- that conflict with the proposed therapeutic mechanisms of
bilitation; Splints. splints. Presumably, splints are offered to patients to provide
© 2005 by the American Congress of Rehabilitation Medi- short-term support to weakened or sore muscles, to reduce
cine and the American Academy of Physical Medicine and pain, and to cushion the injured area—thereby promoting re-
Rehabilitation turn to activity. But in a randomized study24 of patients with
unilateral epicondylitis, none of the bracing conditions (includ-
PPER-EXTREMITY DISORDERS believed to result ing no brace) was associated with significant differences in
U from repetitive and/or forceful work activity are of major
concern in occupational medicine and workers’ compensation.
pain-free grip strength. In another randomized experimental
study of nonimpaired patients,25 researchers found that wearing
According to the US Bureau of Labor Statistics, between 1988 a forearm band increased rather than reduced extensor muscle
and 1992 the number of repeated trauma cases reported in- fatigue, as measured by wrist extension force and peak grip
creased by 144%, compared with a 3% increase for all other isometric force. Such findings suggest that splints might hinder
injuries and illnesses.1,2 By 2001, repeated trauma accounted activity and potentially contribute to deconditioning, contrary
for 4% of the 5.2 million total workplace injuries and illnesses to indications. Most recently, a randomized clinical trial26 of
and 65% of occupational illnesses, with work-related upper- brace-only, PT, and combination treatment found mixed re-
extremity disorders (WRUEDs) accounting for most of these.3 sults, with different conditions achieving superior results on
Epicondylitis is one of the common WRUED diagnoses. different outcomes at 6 weeks and no significant differences
There are conflicting studies on the cause of epicondylitis and, between groups at 26 and 52 weeks.26
Despite the paucity of evidence to suggest that relative rest,
immobilization, or splinting is indicated in the treatment of
epicondylitis, it is common for patients to be restricted from
From Concentra Inc, Addison, TX.
certain activities at work, as well as to be braced or splinted at
No commercial party having a direct financial interest in the results of the research least initially, to rest the wrist and/or elbow. Such treatment is
supporting this article has or will confer a benefit upon the authors(s) or upon any prescribed even when the job activities are not necessarily
organization with which the author(s) is/are associated. forceful or excessive, which raises the concern of overuse or
Reprint requests to Jenny N. Devenport, PhD, Concentra Inc, 5080 Spectrum Dr,
Ste 400, West Tower, Addison, TX 75001, e-mail: Jenny_Novotny@concentra.com.
misuse of a medical treatment for a condition that might
0003-9993/05/8606-9367$30.00/0 resolve more quickly without restrictions, rest, or immobiliza-
doi:10.1016/j.apmr.2004.11.029 tion in many cases.20 It is quite possible, for example, that once

Arch Phys Med Rehabil Vol 86, June 2005


1082 SPLINTING AND EPICONDYLITIS OUTCOMES, Derebery

Table 1: Initial Differences Between Treatment Groups, Splint Versus No Splint


Splint No Splint Comparisons†
Mean ⫾ SD Mean ⫾ SD Standardized
Variable (n⫽3236) (n⫽1378) Diff in %‡

% with high severity ¶


89.7⫾30.4 84.0⫾36.7 ⫺16.95*
Patient age (y) 41.42⫾9.29 41.32⫾9.54 ⫺1.00
Sex (% male) 57.8⫾49.4 61.2⫾48.7 7.12*
Marital status (% married)储 50.1⫾50.0 47.8⫾50.0 ⫺4.66
% with lateral epicondylitis 82.4⫾38.1 78.3⫾41.2 ⫺9.38*
% treated within 1wk of injury 59.9⫾49.0 55.2⫾49.7 ⫺10.16*
% initial provider⫽physician 71.9⫾44.9 73.3⫾44.3 3.04
% with prior case 37.2⫾48.3 35.1⫾47.7 ⫺4.42
% with initial drug screen 26.3⫾44.0 22.0⫾41.4 ⫺10.15*
2
Region (4: midwest/east, south, west, southeast) ␹3 test⫽36.21 *

Abbreviation: SD, standard deviation.


*P⬍.05.

The Fisher exact test used to compare means for all variables except age and region; t test used for age; chi-square used for region.

The standardized difference is the mean difference between groups divided by the average standard deviation: 100(ms⫺mns)/公[(ss2 ⫹ sns 2
)/2], where for
each covariate ms and mns are the subclass means for the splinted and not splinted cases, respectively, and ss2 and sns
2
are the corresponding variances.

Internal rating given by initial treating provider (original scale: no objective findings, mild, moderate, severe; used moderate ⫹ severe vs mild
⫹ no objective findings based on frequency analysis, suggesting dichotomous variable more reflective of use).

N⫽3219 and 1376 for splinted and not splinted, respectively, because of missing data.

a patient becomes symptomatic of epicondylitis, certain routine year 2002 and ending by July 31, 2003, were eligible for the
activities cause an increase in symptoms, misleading the pa- study. Fifty-one patients were excluded because of concurrent
tient (and doctor) to think that the activity “caused” the disor- treatment for complicating injuries not involving the elbow. A
der and should be avoided even when use of the arm, despite total of 4614 patients remained for inclusion in subsequent
pain, may in actuality lead to a more rapid recovery. analyses.
The primary goal of treatment of epicondylitis should be the
patient’s rapid and enduring return to full functioning. For Splinting
splinting to be recommended in the context of evidence-based Splinting was identified from the electronic records for pa-
medicine, it would be prudent to show that patients with splints tient visits. A variety of restraints was provided to these pa-
show faster and more sustained improvements in pain and/or tients, at their treating provider’s discretion, to reduce move-
function than patients without splints. Specific to occupational ment and apply supportive tension to the elbow/wrist and
medicine and workers’ compensation, patients with splints associated extensor muscles. Because the comparison of inter-
would need to have less time off work, less duty restrictions, est in this study was splints versus no splints (rather than
shorter treatment durations, and lower medical costs than pa- comparisons of splints applied to the elbow versus both the
tients without splints. To date, we know of no studies that elbow and wrist), patients were counted as receiving a splint if
address each of these work-related outcomes relative to splint- they received any restraint to the elbow, forearm, or wrist
ing. areas—including braces, splints, straps, and wrap bandages.
The purpose of this article was to evaluate the effects of
splinting on outcomes for injured workers with epicondylitis. Evaluation of Splinting Differences
Specifically, retrospective analyses of patients were undertaken To control for pretreatment differences between patients
to identify pretreatment differences in splinted and nonsplinted who did and did not receive splints, the first step was to identify
patients on background and initial injury characteristics, to and quantify observed differences. Table 1 presents summary
statistically control for these differences, and to estimate the information on available background characteristics and initial
effects of splinting on functional and treatment outcomes most condition for patients who did and did not receive splints.
relevant to occupational medicine. Independent t tests and chi-square tests were conducted to
assess group differences. No significant differences were ob-
METHODS served for patient age (measured in years), the type of initial
treating provider (primary care physician [MD or DO] vs
Participants physician assistant), patient marital status (single, married),
The patient population in this study consisted of patients and the existence of any prior injury claim within the network
receiving primary care for lateral or medial epicondylitis (In- (yes, no). Significant differences were observed for severity
ternational Classification of Diseases, 9th Revision, codes ratings by the initial treating provider (patients with moderate
726.31 or 726.32) at any clinic within a nationwide network of to severe ratings were more likely to receive splints than
253 occupational medical centers. This network, owned and patients with mild ratings), gender (higher rates of splinting for
operated by Concentra Inc, sees approximately 7% of US females versus males), diagnosis (higher rates of splinting for
workers’ compensation patients. Patient records were retrieved patients with lateral versus medial epicondylitis), for treatment
from the proprietary internal information management system, lag (higher rates of splinting for patients treated within 1wk of
which contains patient demographic and injury information, as injury vs those treated ⬎1wk), for drug screen (higher rate of
well as transaction-level treatment, diagnostic, billing, and splinting for patients who received a drug screen, an optional
outcomes information. All patients who were of legal working service provided during the initial visit at the discretion of the
age (at least 16 years old) and who received primary injury care employer or governing laws of the area), and for geographic
at the centers with treatment episodes beginning in calendar region (patients in the west and southeast regions received

Arch Phys Med Rehabil Vol 86, June 2005


SPLINTING AND EPICONDYLITIS OUTCOMES, Derebery 1083

splints more than patients in the south, midwest, and east score subclasses, there were no remaining statistically signifi-
regions). cant differences between splinted and nonsplinted patients on
any of the covariates. Figure 1 displays side-by-side boxplots
Statistical Control of Pretreatment Differences of the propensity scores for splinted and nonsplinted patients
In a retrospective study of usual care patterns, it is not within each of the 5 subclasses. The degree of symmetry in the
reasonable to presume that splinting decisions were made at medians, the similarity in the size of the boxes, and the length
random, that patients had an equal chance of having a splint of the whiskers further support the adequacy of the final pro-
versus not having a splint, or that pretreatment differences pensity score model for reducing pretreatment differences on
among patients were uncorrelated with outcome prognoses. In observed covariates.
our example, patients with more severely rated injuries were Finally, to estimate outcome differences between splinted
more likely to receive splints, but we might also expect patients and nonsplinted patients while controlling for background co-
with more severe injuries to miss work, to require more duty variates, the within-subclass means and standard errors (SEs)
restrictions, to incur more medical charges, and to have longer for the 2 groups were calculated and then used to compute
treatment durations than less severe cases even before splinting overall differences and standard errors.28,33 The outcomes of
decisions. Thus, an immediate comparison of these outcomes interest were the percentage of cases put on limited duty or
for patients with and without splints would be misleading. taken off work by the treating provider (as distinct from the
Propensity score methodology is a statistical approach used percentage of patients who elect to take off work or are taken
in observational studies to control simultaneously for multiple off work by their employers), treatment duration (from the first
pretreatment covariates to create equivalent groups of patients date of service to the last date of service), the percentage of
who did and did not receive a particular treatment.27-31 The patients who completed their care in the network (eg, released
propensity score represents the conditional probability of re- from care to full duty by the network vs opting out), and the
ceiving a particular treatment—in our study, a splint— given percentage of paients referred to specialists. Limited duty rates,
observed covariates. Simply put, the propensity score reduces lost time rates, and treatment durations are standard measures
the differences between patients on several variables to a of the effectiveness of medical management in workers’ com-
one-number summary. If 2 patients, one who received a splint pensation cases. Treatments that minimize these measures
and one who did not, have the same propensity to receive a while improving patient function and symptoms are preferred.
splint (predicted from a set of potential confounding variables), Referral to a specialist generally indicates the failure of con-
then they would not systematically differ with respect to the servative treatment efforts to yield adequate improvement. The
predictor variables.29,31 That is, these variables would not help final outcome, completion rate, is potentially indicative of
predict which of these 2 patients received the splint. As in a patient and/or employer satisfaction with care, in addition to
randomized experiment, the treatment assignment would be success of care. If large disparities in completion rates exist
independent of these predictor variables. Thus, observed dif- between treatment groups, the treatment with the higher com-
ferences in outcomes for these patients could not be attributed pletion rate would be preferred (assuming the treatment is
to the pretreatment differences on the predictors because they effective).
were balanced at the start. Similarly, if patients are divided into In addition to the main hypothesis, splinting versus not
equally sized subclasses based on the magnitude of their pro- splinting, the effect of PT was also assessed. The medical
pensity scores, it has been shown that the average treatment model of the network in our study favors aggressive therapeutic
effect within subclasses (the difference between splinting and interventions—including strength training, patient education,
not splinting within a given range of propensity scores) will be manual therapy, and electrotherapeutic modalities—so the
an unbiased estimate of the true treatment effect.27-28,32 splinting effect needed to be measured in the presence or
absence of PT interventions. Propensity scores for splinting
were reestimated separately for patients who did and did not
Estimating Propensity Scores and Treatment Effects receive PT. For patients who received therapy, the lag from
In our study, the propensity to receive a splint was estimated initiation of the episode of care to the initiation of PT was also
by using a logistic regression model with splint (yes, no) as the included as a covariate in the model. Models were estimated
dichotomous dependent variable and the demographic and pre- and evaluated for adequacy of covariate and overall score
treatment covariates described earlier as predictors. Modeling balance by using the methods described earlier for the main
proceeded iteratively. Once a solution was estimated, the re- comparison. Not all the same main effects and interactions
sulting propensity scores (conditional probability of receiving a were included in these models, although balance on all the
splint) were divided into 5 equally sized subclasses based on covariates and the resulting estimated propensity scores were
their rank (ie, patients with the lowest propensity scores in the verified. The within-subclass means and SEs were then calcu-
first subclass, patients with the highest propensity scores in the lated and used to assess the splinting effect for patients with
fifth subclass). Then the success of the model at achieving and without PT.
balance on the covariates was evaluated by looking at within-
and across-subclass differences between splinted and non- RESULTS
splinted patients for each covariate (using analysis of variance
or logistic regression as appropriate). The initial main effects
model that included all the covariates alone (and no interac- Splinting Main Effect
tions) did not result in balance— defined as no significant main Table 3 summarizes differences in outcome measures for
effects for splinting or interaction effects of splinting and patients with and without splints. Means and SEs for each
subclass on all covariates. This model was refined by adding outcome measure are presented for splinted versus nonsplinted
clinically reasonable and/or statistically significant covariate patients within propensity score subclasses and averaged across
interactions. The final model, presented in table 2, includes all subclasses. Significance test results, noted in the last row of the
significant main effects, plus the main effects used in the table, refer to the overall mean difference observed for splint-
included interactions. By using this model, balance was ing versus not splinting (calculated as a 2-tailed z test, signif-
achieved on all covariates, meaning that, within propensity icant at P⬍.05).

Arch Phys Med Rehabil Vol 86, June 2005


1084 SPLINTING AND EPICONDYLITIS OUTCOMES, Derebery

Table 2: Logistic Regression Model Estimates for Propensity to Receive Splint


Covariate Estimate SE Wald z Score df P

% with high severity (SEV) .69 .13 26.96 1 .00


Patient age (AGE) .00 .00 0.40 1 .53
Marital status (MSAT) .38 .18 4.64 1 .03
Gender (% male) ⫺.14 .07 4.41 1 .04
% treated within 1wk of injury .16 .07 5.72 1 .02
% with lateral epicondylitis .26 .08 9.78 1 .00
% initial provider⫽physician (TYPEPROV) ⫺.25 .10 6.18 1 .01
% with prior case (PREVIOUS) .16 .09 3.25 1 .07
REGION 10.88 3 .01
REGION (1) .17 .20 0.76 1 .38
REGION (2) .43 .20 4.48 1 .03
REGION (3) .94 .33 8.13 1 .00
% with initial drug screen (DRUGSCRN) .72 .35 4.18 1 .04
AGE by DRUGSCRN ⫺.01 .01 1.74 1 .19
TYPEPROV by REGION 11.58 3 .01
TYPEPROV by REGION (1) ⫺.59 .24 6.36 1 .01
TYPEPROV by REGION (2) ⫺.01 .23 0.00 1 .97
TYPEPROV by REGION (3) ⫺.88 .36 5.86 1 .02
PREVIOUS by REGION 9.16 3 .03
PREVIOUS by REGION (1) ⫺.74 .30 5.81 1 .02
PREVIOUS by REGION (2) ⫺.80 .35 5.12 1 .02
PREVIOUS by REGION (3) .28 .52 0.30 1 .58
TYPEPROV by REGION by PREVIOUS 6.32 3 .01
TYPEPROV by REGION (1) by PREVIOUS .92 .37 6.06 1 .01
TYPEPROV by REGION (2) by PREVIOUS .48 .40 1.42 1 .23
TYPEPROV by REGION (3) by PREVIOUS .04 .54 0.00 1 .95
MSTAT by SEV ⫺.33 .19 2.93 1 .09
Constant ⫺.02 .24 0.01 1 .94

NOTE. Hosmer and Lemeshow goodness of fit: ␹82 test⫽4.39, P⫽.820. Omnibus model ␹23
2
test⫽140.46, P⬍.001.
Abbreviation: SE, standard error.

Overall significant differences between splint groups were treatment duration. Patients with splints had higher rates of
found for rate of limited duty, number of medical visits, med- limited duty within each subclass and overall (range, 7%–20%)
ical charges, total primary care charges (includes PT), and than nonsplinted patients. Typical duty modifications and re-
strictions prescribed by the treating physician involved lifting
restrictions and limitations on time spent doing forceful or
repetitive tasks. Patients with splints also had an average of 1
more medical visit within and across all subclasses, higher
medical charges (expected given the additional medical visit),
higher total charges (PT plus medical), and treatment durations
that were an average of 12 days longer than patients without
splints. Differences in rates of lost time, the percentage of
patients completing care, and the percentage of patients re-
ferred out to specialists were not statistically significant.
These results do not suggest an advantage for splinting from
an outcomes perspective. With the adjustments included for
pretreatment differences, these results suggest that splinted
patients had worse outcomes in terms of treatment duration,
return to activity, and medical costs without improving rates of
care completion than did patients without splints. However, the
significant differences in total charges (which include therapy
charges) do suggest a need for further investigation of the
hypothesis that therapy interventions contribute to higher costs
and longer treatment durations.

Splinting and Therapy


Table 4 exhibits outcome differences for splinting and not
splinting in patients who did not receive PT. Means with SEs
are presented within and across subgroups for each outcome
Fig 1. Boxplot showing balance of propensity scores within sub- measure. Overall significance results are presented in the bot-
classes. Confidence interval is 99%. tom row for each outcome variable. Significant differences

Arch Phys Med Rehabil Vol 86, June 2005


SPLINTING AND EPICONDYLITIS OUTCOMES, Derebery 1085

Table 3: Estimated Outcome Differences Between Groups, Splint Versus No Splint*


%
Completed
Care (full
No. of Avg MD Avg MD Avg Total duty Avg Treatment % Refd Out
Subclass Splint Patients % Ltd Duty % Lost Time Visits Charges ($) Charges ($) release) Duration to Specialist

1 No 383 48.3⫾2.6 1.3⫾0.6 2.88⫾0.12 321.21⫾13.50 709.20⫾35.74 53.0⫾2.6 24.31⫾2.14 18.0⫾2.0


Yes 536 68.5⫾2.0 0.9⫾0.4 4.03⫾0.11 459.38⫾13.06 1108.39⫾41.77 56.7⫾2.1 36.11⫾2.36 22.0⫾1.8
2 No 305 60.3⫾2.8 1.6⫾0.7 3.46⫾0.15 406.60⫾20.70 899.82⫾53.30 56.0⫾2.8 30.66⫾3.31 16.4⫾2.1
Yes 614 76.1⫾1.7 1.1⫾0.4 4.37⫾0.10 510.48⫾13.57 1205.58⫾36.97 57.8⫾2.0 41.73⫾2.73 21.2⫾1.6
3 No 283 62.5⫾2.9 1.8⫾0.8 3.42⫾0.14 413.35⫾16.30 843.41⫾40.33 54.1⫾3.0 35.05⫾3.55 20.5⫾2.4
Yes 636 79.9⫾1.6 1.4⫾0.5 4.56⫾0.12 563.38⫾18.74 1285.71⫾47.53 59.0⫾2.0 40.63⫾2.57 20.1⫾1.6
4 No 226 74.3⫾2.9 0.9⫾0.6 3.72⫾0.17 449.24⫾23.32 895.24⫾49.45 60.2⫾3.3 34.28⫾3.87 17.7⫾2.5
Yes 693 81.5⫾1.5 1.4⫾0.4 4.90⫾0.13 596.57⫾17.28 1283.01⫾39.06 59.7⫾1.9 47.72⫾2.57 26.7⫾1.7
5 No 179 68.2⫾3.5 2.2⫾1.1 3.78⫾0.23 421.28⫾24.89 819.10⫾58.62 60.3⫾3.7 36.26⫾4.75 19.6⫾3.0
Yes 740 79.3⫾1.5 1.6⫾0.5 5.15⫾0.13 606.98⫾17.36 1274.17⫾37.78 62.4⫾1.8 54.38⫾2.89 27.4⫾1.6
Overall No 1376 62.7⫾2.9 1.6⫾0.8 3.45⫾0.16 402.33⫾20.20 833.36⫾48.22 56.7⫾3.1 32.11⫾3.63 18.4⫾2.4
Yes 3219 77.0⫾1.7 1.3⫾0.4 4.60⫾0.12 547.36⫾16.16 1231.37⫾40.80 59.1⫾2.0 44.11⫾2.63 23.5⫾1.7
P⬍.01 NS P⬍.01 P⬍.01 P⬍.01 NS P⬍.01 NS

NOTE. Values mean ⫾ SE unless otherwise indicated. N⫽4595. Marital status, which was included in the propensity score estimation, was not
provided by 19 patients. Thus, their propensities were not estimated and included in this set, although they were included in later analyses.
Abbreviations: Avg, average; Ltd, limited; NS, not significant; Refd, referred.
*Subclasses of equal size were created according to the ranked propensity score; within subclasses, patients with and without splints have
similar propensity scores; overall estimates were produced using direct standardization methods with subclass total weights.34,41

were again observed for rates of limited duty, number of the differences in treatment duration observed in the overall
medical visits, and medical charges. As with the overall results, population. But the consistency of results independent of ther-
outcome differences did not favor splinting. Patients without apy suggests that splinting does not promote, and may even
splints were less likely to have limited duty, had fewer medical impede, optimal outcomes.
visits, and had lower medical charges than their splinted coun-
terparts. However, no differences in treatment duration were ob- DISCUSSION
served, in contrast with the overall results presented in table 3. A treating provider’s main reasons for prescribing a splint
Table 5 presents analogous comparisons for splinting and are, presumably, to rest the arm and to alleviate pain or dis-
not splinting in patients who received PT. Again, statistically comfort. However, no good correlation between a patient’s
significant differences in limited duty rates, medical visit subjective pain ratings and his/her ability to work or to perform
counts, and medical charges were observed, and worse out- certain physical activities has been established, nor has a cor-
comes for splinted patients were shown. In addition, splinted relation been found between the decrease in symptoms and the
patients received an average of 1 more therapy visit than rate of return to work.34,35 In addition, prescribing a splint is
nonsplinted patients and had higher overall charges for therapy. likely to necessitate movement restrictions that may further
The total treatment duration difference was not significant, impede recovery and contribute to disability in those cases in
which suggests that additional therapy, either alone or in com- which the job activities, which may lead to discomfort, do not
bination with additional medical visits, may be responsible for actually cause the condition. Such restrictions can have an

Table 4: Estimated Outcome Differences Between Groups, Splint Versus No Splint for Cases Without PT*
% Completed
No. of Avg MD Charges Care Avg Treatment % Refd Out
Subclass Splint Patients % Ltd Duty % Lost Time Avg MD Visits ($) (full duty release) Duration to Specialist

1 No 154 24.7⫾3.5 1.3⫾0.9 2.14⫾0.14 249.70⫾14.03 49.35⫾4.04 18.18⫾2.80 13.6⫾2.8


Yes 105 39.0⫾4.8 1.0⫾1.0 2.89⫾0.20 348.00⫾24.36 55.24⫾4.88 21.30⫾3.60 10.5⫾3.0
2 No 115 33.9⫾4.4 0.9⫾0.9 2.49⫾0.27 323.54⫾33.26 43.48⫾4.64 22.15⫾4.56 14.8⫾3.3
Yes 143 58.7⫾4.1 0.0⫾0.0 3.31⫾0.18 395.36⫾21.10 51.75⫾4.19 25.17⫾3.25 15.4⫾3.0
3 No 103 37.9⫾4.8 0.0⫾0.0 1.99⫾0.13 265.67⫾16.59 43.69⫾4.91 22.96⫾4.63 11.7⫾3.2
Yes 158 55.7⫾4.0 0.0⫾0.0 3.29⫾0.19 391.21⫾22.78 47.47⫾3.99 27.47⫾3.90 20.3⫾3.2
4 No 87 41.4⫾5.3 1.1⫾1.1 2.40⫾0.20 319.52⫾23.57 48.28⫾5.39 20.90⫾4.14 18.4⫾4.2
Yes 172 58.7⫾3.8 0.6⫾0.6 3.19⫾0.17 418.37⫾21.12 51.74⫾3.82 27.37⫾4.34 15.7⫾2.8
5 No 69 59.4⫾6.0 4.3⫾2.5 2.93⫾0.29 339.54⫾34.44 56.52⫾6.01 20.07⫾5.97 15.9⫾4.4
Yes 190 48.4⫾3.6 0.5⫾0.5 3.12⫾0.20 368.02⫾20.80 54.21⫾3.62 30.49⫾5.00 12.6⫾2.4
Overall No 528 39.4⫾4.9 1.5⫾1.3 2.39⫾0.21 299.52⫾25.75 48.26⫾5.04 20.85⫾4.53 14.9⫾3.6
Yes 768 52.1⫾4.1 0.4⫾0.6 3.16⫾0.19 384.19⫾22.08 52.08⫾4.12 26.36⫾4.06 14.9⫾2.9
P⬍.05 NS P⬍.01 P⬍.01 NS NS NS

NOTE. Values mean ⫾ SE unless otherwise indicated.


*Subclasses of equal size were created according to the ranked propensity score; within subclasses, patients with and without splints have
similar propensity scores; overall estimates were produced using direct standardization methods with subclass total weights.

Arch Phys Med Rehabil Vol 86, June 2005


1086 SPLINTING AND EPICONDYLITIS OUTCOMES, Derebery

adverse impact on the patient, not only physiologically but also

*Subclasses of equal size were created according to the ranked propensity score; within subclasses, patients with and without splints have similar propensity scores; overall estimates
4.89⫾0.26 684.57⫾41.59
5.97⫾0.23 886.53⫾37.18
5.30⫾0.30 707.88⫾41.79
6.46⫾0.25 931.13⫾39.60
5.59⫾0.37 728.77⫾44.20
6.65⫾0.24 946.81⫾36.17
5.44⫾0.34 642.31⫾37.65
6.64⫾0.22 828.20⫾27.27
6.56⫾0.53 735.46⫾56.28
7.65⫾0.27 894.93⫾29.46
5.56⫾0.37 699.85⫾44.76
6.67⫾0.24 897.56⫾34.27
psychologically.36 Splint use sends a powerful message, not

Charges ($)
Avg PT

P⬍.01
only to the patient but also to family and colleagues, that the
patient has an injury and that the injured arm needs to be rested.
Because good physical conditioning is helpful to the healing
process of nontraumatic injuries as well as to job performance
under healthy circumstances, it is all the more important to
Avg PT

P⬍.01
Visits

consider carefully both the risks and the benefits of splinting.


Our study used patient data collected by a large occupational
medicine network to explore the effects of splinting versus not
splinting on outcomes for epicondylitis patients. Pretreatment
% Refd Out to

16.2⫾2.4
21.4⫾2.0
22.2⫾2.9
21.7⫾1.9
22.2⫾3.3
24.7⫾1.9
18.5⫾3.2
27.5⫾2.0
28.0⫾4.4
34.8⫾2.0
21.4⫾3.3
26.0⫾2.0
Specialist

differences in observed demographic and initial injury charac-

NS
teristics were identified and controlled for by using propensity
score methods. The obtained results suggest that the presumed
benefits of splinting— commonly including alleviation of dis-
Table 5: Estimated Outcome Differences Between Groups, Splint Versus No Splint for Cases With PT*

comfort, cushioning of the area, and support of weakened


Avg Treatment

muscles—may not translate to better outcomes. Rather, splint-


29.29⫾3.17
37.26⫾2.70
36.08⫾4.47
47.26⫾3.39
31.83⫾3.23
44.60⫾3.13
44.94⫾6.03
50.01⫾2.58
56.53⫾7.59
68.01⫾3.66
39.74⫾5.19
49.44⫾3.12
Duration

NS ing appears to be related to higher treatment utilization and


costs and higher rates of duty restrictions (which translate to
higher indemnity costs for workers’ compensation cases) than
not splinting— even after adjusting for important pretreatment
differences, such as severity, age, injury history, geographic
% Completed Care
(full duty release)

region, and other characteristics.


The results of our study are consistent with research on
60.0⫾3.2
55.6⫾2.4
61.4⫾3.4
61.9⫾2.3
63.0⫾3.8
62.2⫾2.2
62.3⫾4.0
62.8⫾2.1
57.9⫾4.8
64.7⫾2.0
60.9⫾3.9
61.4⫾2.2

activity restriction reported for other types of musculoskeletal


NS

injuries. Back pain studies have found that patients who remain
active despite back pain do better than those who rest.37,38
Similarly, it has been reported that, in chronic back pain
patients who experience increased pain during the first month
1045.22⫾52.86
1351.29⫾44.40
1171.23⫾57.11
1499.28⫾49.27
1221.17⫾65.18
1545.47⫾54.14
1128.03⫾55.22
1465.46⫾40.42
1279.69⫾86.21
1614.10⫾44.32
1169.16⫾64.49
1495.22⫾46.75

of a strenuous exercise program, if they are told to continue


Charges ($)
Avg Total

P⬍.01

performing the exercises anyway despite pain, they go on to


have better outcomes than patients whose treatments involve
less vigorous activity.39 In another study,40 patients with rheu-
matoid arthritis who were placed on an intensive, dynamic
exercise program had better joint mobility, muscle strength,
360.66⫾20.64
464.76⫾13.89
463.34⫾22.97
568.16⫾17.71
492.40⫾27.81
598.66⫾22.21
485.72⫾24.21
637.26⫾19.23
544.24⫾38.25
719.17⫾22.21
469.32⫾27.49
597.65⫾19.30

and physical conditioning than patients whose treatments were


Charges ($)
Avg MD

P⬍.01

less intensive.40 These studies represent a sample of a growing


were produced using direct standardization methods with subclass total weights.

trend in almost every medical discipline toward early activation


in injury management.
A limitation of this study is its retrospective design, which
restricts certainty about the causal relationship between splint-
Avg MD Visits

3.36⫾0.12
4.19⫾0.11
3.98⫾0.19
4.82⫾0.13
4.14⫾0.18
4.92⫾0.14
4.26⫾0.22
5.28⫾0.14
4.97⫾0.33
6.05⫾0.15
4.14⫾0.22
5.05⫾0.14

ing and outcomes. In a randomized controlled trial (RCT),


P⬍.01

pretreatment differences between patients are randomly allo-


cated among treatment and control groups, so balance on both
observed and unobserved characteristics is assured. Although
propensity score methodology was used here to successfully
balance splinted and nonsplinted patients on all available (ob-
% Lost Time

NOTE. Values mean ⫾ SE unless otherwise indicated.


1.3⫾0.8
1.4⫾0.6
2.4⫾1.0
1.3⫾0.5
1.2⫾0.9
1.2⫾0.5
0.7⫾0.7
1.4⫾0.5
2.8⫾1.6
2.7⫾0.7
1.7⫾1.0
1.6⫾0.6

served) pretreatment differences, unobserved differences, by


NS

definition, were not controlled for. To the extent that an im-


portant predictor of splinting and outcomes was not available
for consideration, and particularly if this omitted characteristic
does not correlate with the variables included, propensity
% Ltd Duty

65.4⫾3.2
77.5⫾2.0
76.3⫾3.0
87.5⫾1.6
81.5⫾3.0
84.9⫾1.6
80.1⫾3.3
87.4⫾1.5
82.2⫾3.7
87.3⫾1.4
77.1⫾3.2
84.9⫾1.6
P⬍.05

scores may not have adequately removed selection bias from


splint allocation (eg, the differences in patient or treatment
conditions that may have influenced both splinting decisions
and outcomes).
Patients

The sensitivity of the splint and no-splint results was eval-


No. of

228
435
207
457
162
502
146
516
107
558
850
2468

uated by looking at patients who received PT separately from


those who did not. The contribution of therapy can be inter-
preted in 2 ways. First, the sports medicine model of this
Splint

Yes

Yes

Yes

Yes

Yes

Yes
No

No

No

No

No

No

occupational network emphasizes early, aggressive therapeutic


intervention to achieve optimal outcomes. Therefore, as far as
differentially pairing treatments for patients with and without
Subclass

Overall

splints, it seems possible that patients without splints would


receive PT sooner. A statistically significant association be-
1

tween splinting and lag to therapy was observed in the study

Arch Phys Med Rehabil Vol 86, June 2005


SPLINTING AND EPICONDYLITIS OUTCOMES, Derebery 1087

population—91% of patients without splints started PT within www.bls.gov/news.release/osh.nr0.htm. Accessed November 29,
a week of beginning care versus 88% with splints. To the 2004.
extent that early therapy makes a difference in outcomes and 4. Kurppa K, Viikari-Juntura E, Kuosma E, Huuskonen M, Kivi P.
correlates with splinting decisions, the observed differences Incidence of tenosynovitis or peritendinitis and epicondylitis in a
could have been incorrectly labeled as splinting effects. Sec- meat processing factory. Scand J Work Environ Health 1991;17:32-7.
ond, the measure of severity used in the propensity score model 5. Viikari-Juntura E, Kurppa K, Kuosma E, et al. Prevalence of
was dichotomous and, as such, might not achieve fine discrim- epicondylitis and elbow pain in the meat-processing industry.
inations between patients of differing injury severity. But pa- Scand J Work Environ Health 1991;17:38-45.
tients with mild severity were less likely to be referred to a 6. Chiang HC, Ko YC, Chen SS, Yu HS, Wu TN, Chang PY.
physical therapist for supervised exercise sessions or applica- Prevalence of shoulder upper-limb disorders among workers in the
tion of modalities and were more likely to be given instructions fish-processing industry. Scand J Work Environ Health 1993;19:
on a home exercise regimen. The fact that differences between
126-31.
splinted and nonsplinted patients persisted independent of re-
7. Byström S, Hall C, Welander T, Kilbom A. Clinical disorders and
ceipt of therapy services offers some evidence that the differ-
ences have been appropriately labeled and that reduction of pressure pain threshold of the forearm and hand among automo-
selection bias because of severity has occurred. bile assembly line workers. J Hand Surg [Br] 1995;20:782-90.
Another limitation of this study is that differences between 8. Ritz BR. Humeral epicondylitis among gas and waterworks em-
types of splints were not assessed. The main hypothesis of this ployees. Scand J Work Environ Health 1995;21:478-86.
study was that any type of splint involves restriction of move- 9. Kurvers H, Verhaar J. The results of operative treatment of medial
ment and would result in longer treatment durations, higher epicondylitis. J Bone Joint Surg Am 1995;77:1374-9.
costs, and more time away from work than no splint at all. By 10. Moore J, Garg A. Upper extremity disorders in a pork processing
extension, under this hypothesis, splints creating greater re- plant: relationships between job risk factors and morbidity. Am
striction of movement and worn for longer periods of time Ind Hygiene Assoc J 1994;55:703-15.
might result in worse outcomes than splints with less restriction 11. Tichauer E. Biomechanics sustains occupational safety and health.
of movement worn for shorter periods of time. These questions Ind Eng 1976;8:44-5.
were beyond the scope of our analysis but would make for 12. Descatha A, Leclerc A, Chastang F, Roquelaure Y. Medical
informative future follow-up investigations. epicondylitis in occupational settings: prevalence, incidence and
associated risk factors. J Occup Environ Med 2003;45:993-1001.
CONCLUSIONS 13. Dimberg L, Olafsson A, Stefansson E, et al. The correlation
The challenge in occupational medicine, particularly with between work environment and the occurrence of cervicobrachial
respect to workers’ compensation, is for providers to maximize symptoms. J Occup Med 1989;31:447-53.
the health and well-being of their patients while showing their 14. National Institute for Occupational Safety and Health. Bernard
cost-effectiveness to employers and payers in the present en- BP, editor. Musculoskeletal disorders and workplace factors: a
vironment of ever-escalating medical costs. Critical to the critical review of epidemiologic evidence for work-related mus-
success of such efforts is an evaluation of “what works” be- culoskeletal disorders of the neck, the upper-limb, and low back.
cause cheaper procedures are not cost-effective if they inflate 2nd printing. Cincinnati: NIOSH; 1997.
total medical and indemnity costs by prolonging treatment 15. Jackson M. Evaluating and managing tennis elbow. Your Patient
duration and increasing duty restrictions and time off work.
Fitness 1997;11(2):104i-104l.
The findings of our study provide evidence that splinting
16. Chard M, Hazleman B. Tennis elbow: a reappraisal. Br J Rheu-
epicondylitis patients does not necessarily lead to better out-
comes and, in fact, may have adverse effects. Splinted patients matol 1989;28:186-90.
had higher rates of limited duty, more medical visits, higher 17. Jobe F, Ciccotti M. Lateral and medical epicondylitis of the
medical costs, and longer treatment durations than similar elbow. J Am Acad Orthop Surg 1994;2(1):1-8.
patients without splints. Nonetheless, these results should not 18. Hay E, Paterson S, Lewis M, Hosie G, Croft P. Pragmatic ran-
replace high-quality RCTs of splinting; rather, they should domized controlled trial of local corticosteroid injection and
encourage their conduct, to determine under what conditions or naproxen for treatment of lateral epicondylitis of the elbow in
for which patients, if any, splinting produces better outcomes, primary care. BMJ 1999;319:964-8.
including long-term function and productivity. Additional stud- 19. Hudak I, Cole D, Haines T. Understanding prognosis to improve
ies might also address what treatment approaches tend to be rehabilitation: the example of lateral elbow pain. Arch Phys Med
paired with splinting versus not splinting in practice, to verify Rehabil 1996;77:586-93.
that the effect observed is correctly attributed to splinting.41 20. Smidt N, van der Windt D, Assendelft W. Corticosteroid injec-
Finally, the effects of splinting decisions on actual versus tions, physiotherapy, or a wait-and-see policy for lateral epicon-
prescribed absences from work, objective measures of patient dylitis: a randomized controlled trial. Lancet 2002;359:657-62.
function, recurrence of symptoms, and patient satisfaction with 21. Kivi P. The etiology and conservative treatment of humeral epi-
treatment should be assessed. condylitis. Scand J Rehabil Med 1988;15:37-41.
22. Verhaar J, Walenkamp G, van Mameren H, Kester AD, van der
References Linden AJ. Local corticosteroid injection vs Cyriax-type physio-
1. US Bureau of Labor Statistics. Repetitive tasks loosen some therapy for tennis elbow. J Bone Joint Surg Br 1996;78:128-32.
workers’ grip on safety and health. Washington (DC): BLS; Aug 23. Hadler N. The “ergonomics injury” as a social construction.
1994. Available at: http://www.bls.gov/iif/oshwc/ossm0005.pdf. Workers Compensation Policy Rev 2001;1;5:20-5.
Accessed November 29, 2004. 24. Wuori JL, Overend TJ, Kramer JF, MacDermid J. Strength and
2. Gerr F, Letz R, Landrigan P. Upper-extremity musculoskeletal pain measures associated with lateral epicondylitis bracing. Arch
disorders of occupational origin. Annu Rev Public Health 1991; Phys Med Rehabil 1998;79:832-7.
12:543-66. 25. Knebel PT, Avery DW, Gebhardt TL, et al. Effects of the forearm
3. US Bureau of Labor Statistics. Workplace injuries and illnesses in support band on wrist extensor muscle fatigue. J Orthop Sports
2001. Washington (DC): BLS; Dec 2002. Available at: http:// Phys Ther 1999;29:677-85.

Arch Phys Med Rehabil Vol 86, June 2005


1088 SPLINTING AND EPICONDYLITIS OUTCOMES, Derebery

26. Struijs PA, Kerkhoffs GM, Assendelft WJ, Van Dijk CN. Con- 34. Ohnmeiss DD, Vanharanta H, Estlander AM, Jamsen A. The
servative treatment of lateral epicondylitis: brace versus physical relationship of disability (Oswestry) and pain drawings to func-
therapy or a combination of both—a randomized clinical trial. tional testing. Eur Spine J 2000;9:208-12.
Am J Sports Med 2004;32:462-9. 35. McCracken LM, Gross RT, Eccleston C. Multimethod assessment
27. Rosenbaum PR, Rubin DB. The central role of the propensity of treatment process in chronic low back pain: comparison of
score in observational studies for causal effects. Biometrika 1983; reported pain-related anxiety with directly measured physical ca-
70:41-55. pacity. Behav Res Ther 2002;40:585-94.
28. Rosenbaum PR, Rubin DB. Reducing bias in observational studies 36. Hall H, McIntosh G, Melles T, Holowachuk B, Wai E. Effect of
using subclassification on the propensity score. J Am Stat Assoc discharge recommendations on outcome. Spine 1994;19:2033-7.
1984;79:516-24. 37. Taimela S, Diederich C, Hubsch M, Heinricy M. The role of
29. Rubin DB. Estimation from nonrandomized treatment compari- physical exercise and inactivity in pain recurrence and absen-
sons using subclassification on propensity scores. In: Proceedings teeism from work after active outpatient rehabilitation for
of the International Conference on Nonrandomized Comparative recurrent or chronic low back pain: a follow-up study. Spine
Clinical Studies; 1997 April 10-11; Heidelberg (Germany). 2000;25:1809-16.
30. D’Agostino RB. Tutorial in biostatistics: propensity score meth- 38. Carragee E, Han M, Yang B, Kim DH, Kraemer H, Billys J.
ods for bias reduction in the comparison of a treatment to a Activity restrictions after lumbar discectomy: a prospective study
non-randomized control group. Stat Med 1998;17:2265-81. of outcomes in 152 cases with no postoperative restrictions. Spine
31. Braitman LE, Rosenbaum PR. Rare outcomes, common treat- 1999;24:2346-51.
ments: analytic strategies using propensity scores. Ann Intern Med 39. Manniche C, Hesselsoe G, Bentzen L, Christensen I, Lundberg E.
2002;137:693-5. Clinical trial of intensive muscle training for chronic back pain.
32. Cochran WG. The effectiveness of adjustment by subclassification Lancet 1988;12:1473-6.
in removing bias in observational studies. Biometrics 1968;24: 40. Intensive exercise for RA. Back Letter 1996;11(1):2.
205-13. 41. Reichardt CS. A typology of strategies for ruling out threats to
33. Mosteller CF, Tukey JW. Data analysis and regression. Reading: validity. In: Brickman L, editor. Research design: Donald Camp-
Addison-Wesley; 1977. p 221-57. bell’s legacy. Vol 2. Thousand Oaks: Sage; 2000. p 89-115.

Arch Phys Med Rehabil Vol 86, June 2005

You might also like