SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. STATISTICAL METHODS
  6. Results
  7. DISCUSSION
  8. Acknowledgements
  9. Conflict of interest
  10. References

Objectives

To compare intra-articular morphine or bupivacaine against no treatment following unilateral elbow joint arthroscopy using force plate analysis and pain scoring.

Methods

Thirty-one dogs were randomly allocated to receive 0·1 mg/kg intra-articular morphine, 0·5 mg/kg bupivacaine or no treatment following elbow arthroscopy. Force plate analysis, pain scoring and kinematic evaluation were performed before anaesthesia, 4 and 24 hours after surgery. Peak vertical force index, symmetry index, rate of loading, rate of unloading, stance time and range of motion were obtained from gait analysis. Pain scoring was performed every 4 hours and interventional analgesia (0·3 mg/kg methadone) was administered if necessary.

Results

Of 29 dogs analysed, peak vertical force index (P<0·001), symmetry index (P=0·01) and rate of unloading (P=0·01) decreased significantly over time in each group; however, this was not affected by treatment. No significant differences were observed in stance time or rate of loading over time. Kinematic (range of motion) evaluation was not complete for all dogs. Pain scores increased significantly at both time points postoperatively in the no treatment group (P=0·007) and in morphine-treated dogs at 4 hours compared to baseline (P=0·03). For intra-articular bupivacaine significant increases in pain scores were not detected (P=0·28).

Clinical Significance

No benefit to intra-articular bupivacaine or morphine was detected using peak vertical force index from force plate analysis. Bupivacaine prevented increases in pain scores at both time points as did morphine at the 24-hour evaluation, compared to no treatment.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. STATISTICAL METHODS
  6. Results
  7. DISCUSSION
  8. Acknowledgements
  9. Conflict of interest
  10. References

Postoperative intra-articular (IA) analgesia for dogs undergoing surgery for cranial cruciate ligament rupture has been extensively studied (Day and others 1995, Sammarco and others 1996, Hoelzler and others 2005). However, similar studies following elbow joint surgery are lacking. Canine elbow dysplasia is a common developmental disorder of the cubital joint of the dog (Houlton 2009) associated with conditions such as fragmentation/fissuring of the medial coronoid process of the ulna.

IA bupivacaine provided the lowest pain scores and higher force threshold after cranial cruciate ligament reconstruction when compared to morphine or saline in dogs (Sammarco and others 1996). In the medical literature IA bupivacaine provides better analgesia than IA morphine in the immediate postoperative period following knee arthroscopy and morphine results in lower pain scores at 24 and 48 hours following surgery (Karlsson and others 1995).

Force plate analysis and kinematic gait evaluation provide objective data on lameness and gait in dogs (Breur and Kim 2008). Kinematic data collection allows accurate description of gait movements whilst force plate analysis allows measurement of forces that affect motion (McLaughlin 2001). These methods of gait analysis have been used as the gold standard to detect differences not observed with subjective lameness scoring. Poor agreement between subjective scoring and objective measures such as gait analysis and large variations between observers using subjective methods have been reported (Quinn and others 2007, Waxman and others 2008). Therefore, it has been suggested to use gait analysis as an objective measurement of lameness in randomised controlled clinical trials where lameness is a primary outcome measure (Breur and Kim 2008).

The objective of this study was to compare IA morphine or bupivacaine to no IA treatment following elbow arthroscopy using force plate analysis, with peak vertical force index (PVFi) and pain scoring as the main outcome measures. It was also intended to document secondary outcome measures derived from force plate analysis [symmetry index (SI), stance time (ST), rate of load and unload] as well as range of motion (ROM) of the elbow joint, using kinematic gait evaluation.

It was hypothesized that an increase in PVFi and a decrease in pain scores would be observed immediately postoperatively with bupivacaine and in the later postoperative period with morphine.

Materials and Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. STATISTICAL METHODS
  6. Results
  7. DISCUSSION
  8. Acknowledgements
  9. Conflict of interest
  10. References

The institutional ethics committee approved this study and informed client consent was obtained.

Dogs with unilateral thoracic limb lameness admitted for unilateral elbow joint arthroscopy were randomly allocated to receive either morphine, bupivacaine or no treatment IA following closure of arthroscopy portals. Dogs with bilateral lameness or receiving concurrent analgesics were excluded. Randomisation was carried out using the envelope method. The dogs were evaluated within 24 hours before premedication and 4 and 24 hours after surgery. Pain scores [Short Form of the Glasgow Composite Pain Scale (SFGCPS); Reid and others 2007] were assigned before each gait analysis session by the investigator who then trotted the dog for gait analysis.

Kinematic and force plate data were collected from both thoracic limbs at trotting speed as described by Hercock and others (2009). In brief, flat reflective markers 10 mm2 were applied to the skin overlying the shoulder, elbow and carpal joints. Positioning was confirmed by palpation of bony landmarks and manual flexion and extension of the limb. The force platform (Kistler Instrument Corp, USA) was located centrally in the middle of a 10×0·5 m runway, flush with the surface. The runway and force platform were covered with a rubber mat. Four Qualysis MCU-500 cameras and Qualysis QTM 2.4 software (Qualysis Medical AB, Sweden) were used to collect kinematic data arranged in a semicircle on one side of the runway at 0·8 to 1·0 m height to ensure that reflective markers were captured by four cameras at a time. Each trial was recorded with a synchronised digital video camera (DCR-PC101E; Sony Corporation, Japan). Before each session the kinematic frame was calibrated to enable marker tracking through the stance phase of the thoracic limb in the recording area.

Dogs were accustomed to the runway before evaluation. A handler trotted the dog along the runway until five valid trials were obtained for each thoracic limb. Trials were considered valid when the lame limb landed in the centre of the force platform within the area bounded by the force transducers at a trotting gait. Trials were considered invalid if the dog was pulling on the lead, not trotting or if the contralateral limb struck the force platform.

Data from the first five valid trials were used for analysis:

  • PVFi – peak vertical force (PVF) divided by body weight
  • SI – PVFi lame limb divided by PVFi contralateral limb.
  • Rate of loading (ROL) – PVF divided by time taken to reach PVF (load the limb).
  • Rate of unloading (ROUL) – PVF divided by time taken to unload the limb.
  • ST – time the lame limb was in contact with the force platform.
  • ROM – the difference in elbow joint angle from flexion to extension

Videos of each trial were reviewed by an experienced veterinary surgeon who provided a lameness score (LNRS) at each evaluation point (0=not lame, 10=very lame).

Preanaesthetic medication consisted of 0·03 mg/kg acepromazine (maximum 1 mg per dog) (ACP injection; Novartis Animal Health) and 0·3 mg/kg methadone (Methadone; Martindale Pharmaceuticals) by intramuscular (im) injection 45 minutes before anaesthesia induction. Propofol (Propoflo; Abbott Animal Health) was administered via an intravenous (iv) cannula slowly to effect. The trachea was intubated with an appropriately sized cuffed endotracheal tube and anaesthesia maintained using isoflurane (Isoflo; Abbott Animal Health) in oxygen delivered by an appropriately sized breathing system. Intravenous fluid therapy with Hartmann's solution (Aqupharm No 11; Animalcare) was provided at 5 mL/kg/hour. Meloxicam (Metacam; Boehringer Ingelheim) 0·2 mg/kg was administered iv. Potentiated amoxicillin (Augmentin; GSK) 20 mg/kg was administered iv every 90 minutes during surgery. If a sympathetic response to surgery was observed (heart rate, blood pressure or respiratory rate increasing by 20% of values obtained before commencement of surgery), 2 µg/kg fentanyl (Fentanyl citrate; Martindale Pharmaceuticals) was administered iv.

Unilateral elbow arthroscopy was performed with standard medial portals (egress, arthroscopic and instrument) as described by Beale and others (2003). Following closure of arthroscopy portals the IA medication, allocated at random, was injected into the elbow joint by the surgeon. In the “no treatment” group nothing was injected into the joint. Volume of injection was 0·1 mL/kg bodyweight, as described by Sammarco and others (1996). To achieve this, 0·1 mg/kg preservative-free morphine (Morphine sulphate; Martindale Pharmaceuticals) was diluted in sterile saline (Vetivex; Dechra) before injection. Bupivacaine 0·5% (Marcain 0·5%; Astra Zeneca) at a dose of 0·5 mg/kg required no dilution.

Pain scoring was performed every 4 hours for the next 24 hours with 0·3 mg/kg methadone administered im if the pain score was above 6/24 or if the observer considered the dog to be painful. The person performing gait analysis and assessing pain scores was unaware of the treatment group.

STATISTICAL METHODS

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. STATISTICAL METHODS
  6. Results
  7. DISCUSSION
  8. Acknowledgements
  9. Conflict of interest
  10. References

Sample size calculations indicated that group sizes of 10 dogs per group would give a statistical power of 0·84 at detecting a difference in PVFi with an alpha of 0·05 using measurements of PVFi in lame (8·7 ±1·5 N/kg) and sound (10·1 ±1·3 N/kg) dogs (Hercock and others 2009). This was based on the assumption that IA analgesics would produce PVFi similar to sound dogs.

Data were entered into Excel Spreadsheets, checked and transferred for statistical analysis into Minitab16 (Minitab Inc., State College, PA, USA), STATA11 and STATA12 (Statacorp, College Station, TX, USA). Basic descriptive statistics were performed and data checked to ensure that the necessary distributional assumptions were fulfilled. Values for age, body mass and gender were compared between groups to verify the randomisation process. General linear models were used to examine the effects on PVFi, ROL, ROUL and ST of time, treatment and the time-treatment interaction. These were supplemented by mixed-effect linear regression using the XTMIXED command in STATA with individual animal being the random effect and using maximum likelihood estimation. The usefulness of competing models was tested using changes in the deviance and individual coefficients within a model were assessed by their Wald statistics.

Differences between SI were examined using a Kruskall-Wallis test as well as an analysis of variance (ANOVA) to look at the effects of time, animal and treatment and time-treatment interaction. A mixed linear regression effect with animal as the random effect examined the effects of treatment and time and their interactions.

For LNRS, statistical significance was investigated using two-way ANOVA with time and treatment as variables. Pain scores were analysed using a Kruskall-Wallis test with post hoc analysis. The relationship between PVFi and the LNRS score was examined with scatter plots and simple linear regression. Statistical significance was taken as P<0·05.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. STATISTICAL METHODS
  6. Results
  7. DISCUSSION
  8. Acknowledgements
  9. Conflict of interest
  10. References

Thirty-one dogs were recruited with 29 included in the data analysis. There were no significant differences with respect to age (P=0·23) or body mass (P=0·74). All dogs demonstrated unilateral thoracic limb lameness with a mean PVFi ±sd of 7·77 ±1·75 N/kg for the lame limb and 10·57 ±1·39 N/kg for the contralateral limb. Males were over-represented, making up 70% of the study population in each group (Table 1).

Table 1. Demographics and arthroscopic diagnosis for 29 dogs undergoing unilateral elbow joint arthroscopy
 No treatment (n=9)Bupivacaine (n=10)Morphine (n=10)P-value
  1. FMCD Fragmented medial coronoid disease, MCD Medial coronoid disease, OA Osteoarthritis

Age, months, mean (sd)35 (18)29 (29)19 (12)0·23
Male:female7:37:37:2 
Body mass, kg, mean (sd)34·2 (6·36)38·7 (11·7)36·2 (19·0)0·74
BreedsLabrador 6

Golden retriever 2

Rhodesian ridgeback 1

Cross 1

Labrador 5

German shepherd 2

Mastiff 1

Flat coat retriever 1

Newfoundland 1

Labrador 5

Akita 1

Shetland sheepdog 1

Cavalier King Charles spaniel 1

Doberman 1

 
Arthroscopic diagnosisFMCD 7

MCD 1

OA 1

FMCD 6

MCD 1

Unknown 3

FMCD 6

MCD 1

OA 1

Chondromalacia 1

Unknown 1

 

Results for the main outcome measures of PVFi and pain scores as well as force-platform-derived parameters were available in all 29 dogs (Table 2).

Table 2. Force-platform-derived parameters from dogs undergoing unilateral elbow joint arthroscopy receiving IA bupivacaine, morphine or no treatment
Parameter/evaluation time (hour)0424Effect
  1. Tx Treatment, TTI Time-treatment interaction

PVFi (N/kg)

Mean/sd

No treatment n=97·67 ±1·946·05 ±1·896·05 ±2·24Time P<0·001

Tx P=0·15

TTI

P=0·6

Bupivacaine n=107·11 ±1·916·79 ±2·706·57 ±1·99
Morphine n=108·52 ±1·176·92 ±1·876·37 ±2·09
SI

Mean/sd

No treatment n=90·71 ±0·180·55 ±0·180·84 ±0·21Time P=0·01

Tx

P=0·48

Bupivacaine n=100·70 ±0·220·60 ±0·260·56 ±0·15
Morphine n=100·81 ±0·130·64 ±0·220·58 ±0·21
Load rate (N/s)

Mean/sd

No treatment n=92024 ±5601602 ±4181414 ±670Time P=0·05

Tx

P=0·66

Bupivacaine

n=10

1797 ±6001719 ±8081647 ±281
Morphine

n=10

1813 ±6921850 ±10581548 ±1224
Unload rate (N/s)

Mean/sd

No treatment n=91915 ±6341428 ±7381318 ±741Time

P=0·01

TTI

P=0·49

Bupivacaine n=101745 ±1141491 ±14271591 ±905
Morphine n=101644 ±13011265 ±9701125 ±816

Results for ROM (Table 3) were only available for 11 dogs because of errors with software picking up the reflective markers and have not been statistically analysed. Pain scores over time are detailed in Table 5.

Table 3. ROM (°) for the lame limb in dogs in three treatment groups before and 4 and 24 hours following unilateral elbow arthroscopy
 Group0 h4 h24 h
ROM (°)No treatment58·8 ±4·1 (n=7)58·6 ±5·5 (n=6)59·3 ±5·8 (n=5)
Mean/sdBupivacaine62·1 ±10·8 (n=5)61·1 ±5·1 (n=3)66·6 ±2·9 (n=4)
 Morphine61·9 ±3·6 (n=6)59·2 ±1·6 (n=5)59·0 ±3·9 (n=3)

A strong negative association was found between PVFi and LNRS (R2=34%, P<0·001). However, scatter plots of this relationship demonstrated a large variation in PVFi for each NRS value and suggest that NRS is not a good predictor of PVFi.

One dog (no treatment group) was excluded, as it could not trot after surgery. Whether this was due to residual sedation from anaesthesia, or pain, was difficult to tell. The pain score was 3/24. Methadone was administered and the dog was removed from further study. Two dogs in the bupivacaine group required one dose of methadone overnight. This was at least 6 hours before gait analysis so were not removed from further evaluation. Data from one dog enrolled (in the morphine group) could not be collected because the dog could not be trained to trot on the force plate.

During surgery fentanyl was required for interventional analgesia in 50% of cases; however, there was no significant difference in fentanyl requirements between groups (P=1·0).

DISCUSSION

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. STATISTICAL METHODS
  6. Results
  7. DISCUSSION
  8. Acknowledgements
  9. Conflict of interest
  10. References

The hypothesis was that bupivacaine would provide an analgesic effect in the immediate postoperative period and morphine in the later postoperative period. The present study demonstrates that an objective measurement of gait and lameness provides no evidence to support this hypothesis, although further studies with a larger sample size would be worthwhile. Pain scoring results do partially support this hypothesis by showing a significant increase in pain scores over time in the no treatment group and no increase in pain scores over time in the bupivacaine group. Our results demonstrate that PVFi, SI and ROUL decrease significantly over time, regardless of treatment, and pain scores increase following unilateral elbow joint arthroscopy.

The PVFi and SI decreased significantly in all groups over time and were not affected by treatment, suggesting that elbow joint arthroscopy is a painful procedure and further work is needed to identify optimal analgesics. These findings are in contrast with Sammarco and others (1996) who demonstrated an immediate postoperative effect of bupivacaine and later effect of morphine, with 12 dogs per group. This could be explained by the current use of an objective measure of lameness (force plate analysis) compared to subjective methods such as pain scoring (McLaughlin and Roush 1995, Jevens and others 1996).

ROL results trend towards an effect of time (slower rate of limb loading following arthroscopy); however, no effect of treatment was noted. The ROUL decreased significantly over time, which suggests that the unloading phase of the gait cycle is more painful and is a parameter worthy of further work in future studies. In dogs with cranial cruciate ligament rupture measures of limb unloading are able to differentiate affected dogs from a normal population (Evans and others 2005).

There was no significant effect of time or treatment on ST in our study. In a study of seven dogs with fragmented medial coronoid disease (FMCD), Burton and others (2008) demonstrated a slightly shortened mean ST in the affected limb (0·24 seconds) compared with the unaffected limb (0·26 seconds). This was in a variety of breeds at trotting speed, similar to our work. In the 29 dogs in our study we compared ST in the lame and contralateral limb and could not demonstrate a significant difference between limbs (Table 4).

Table 4. ST (seconds) for the lame and contralateral limb in dogs in three treatment groups before and 4 and 24 hours following unilateral elbow arthroscopy
 Group0 hoursP-value4 hoursP-value24 hoursP-value
ST (seconds)

Mean/sd

No treatment Lame n=90·27 ±0·040·520·26 ±0·040·170·29 ±0·030·46
No treatment Sound0·26 ±0·02 0·20 ±0·05 0·30 ±0·03 
Bupivacaine Lame n=100·27 ±0·050·910·26 ±0·030·620·26 ±0·030·83
Bupivacaine Sound0·28 ±0·05 0·27 ±0·03 0·26 ±0·04 
Morphine Lame n=100·26 ±0·040·940·26 ±0·040·520·28 ±0·040·68
Morphine Sound0·26 ±0·03 0·28 ±0·07 0·27 ±0·04 

For both ROL and ST, a greater effect may have been noted if the dogs had been walked as opposed to trotted because ST increases with decreased subject velocity (McLaughlin and Roush 1994). This would be worth examining in future studies and results for ROL, ROUL and ST serve as a basis for future sample size calculations.

Subject velocity affects force-platform values and investigations by Riggs and others (1993) recommend that velocity should be tightly controlled in experimental studies. Given the variety of breeds represented in our study and the fact that dogs were recovering from anaesthesia and surgery, implementing and adhering to a standard velocity was deemed difficult. It was therefore decided to trot dogs at a velocity comfortable to the dog. The effect of subject velocity and ST on ground reaction forces has been investigated by McLaughlin and Roush (1994) in walking and trotting Greyhounds. They demonstrated that ST decreased as velocity increased, PVFi increased with increasing velocity and that PVFi decreased as ST increased. They concluded that changes in ST accurately reflect changes in velocity and could be used to normalise trial data. There was no significant alteration in outcome when PVFi data were normalised using ST and therefore we conclude that the variation of subject velocity on PVFi and derived parameters in this study was negligible.

Kinematic evaluation of seven dogs affected with FMCD documented reduced ROM in the affected elbow (Burton and others 2008). A wide reference range of ROM has been described (48·1 to 69·1°) varying with breed, body weight and trotting speed (Clements and others 2005). In the current study baseline data for 18 dogs did not demonstrate a difference in the ROM between the affected (60·8 ±6·3°) and contralateral (59·1 ±6·5°) elbow, which makes using ROM as an outcome measure to compare analgesics difficult. Although data was collected at each time point, group sizes were small because of technical errors and no meaningful conclusions could be drawn (Table 3).

Pain scores (Table 5) were significantly higher at both time points in the no treatment group compared to baseline and in morphine-treated dogs at the 4-hour time point compared to baseline pain scores. Pain scores in the bupivacaine group did not show a statistically significant increase over time, which is in agreement with the work of Sammarco and others (1996). Statistically significant differences in pain scores were recorded during the study although in every case the pain score was lower than the 6/24 intervention level of the SFGCPS (Reid and others 2007). Changes in pain scores over time were small, although any increase in an individual's pain should be viewed as clinically important. For this reason the observer was instructed to administer methadone if they considered the dog to be in pain. Alternatively the ability of the SFGCPS to detect pain in this group of patients can be questioned as there is no doubt that elbow arthroscopy is a painful procedure – a statement which is supported by differences observed in PVFi over time. In this study pain scores increased in the no treatment group following surgery as one would expect and did not increase significantly in the bupivacaine group, supporting an analgesic effect of bupivacaine. The pattern of change in pain scores in the morphine group was as expected from the hypothesis – pain scores increased immediately after surgery but decreased at the 24-hour time point – the time at which previous studies have documented the optimal effect of IA morphine (Karlsson and others 1995).

Table 5. Pain scores over time (SFGCPS) from dogs undergoing unilateral elbow joint arthroscopy receiving IA bupivacaine, morphine or no treatment
  0 hour4 hours24 hours 
  • CI Confidence interval

  • *

    Pain score significantly higher compared to baseline

Pain score (/24)No treatment1·0 [1·0 to 1·0]2·0* [1·47 to 3·0]2·0* [1·47 to 3·0]P=0·007
Median/95% CIBupivacaine1·0 [1·0 to 3·0]2·0 [1·47 to 3·52]3·0 [1·0 to 3·0]P=0·28
 Morphine1·0 [0·47 to 1·0]2·0* [1·0 to 2·0]1·0 [1·0 to 2·0]P=0·03

Two dogs received methadone but were not excluded from further analysis. Both dogs received methadone at least 6 hours before gait analysis. Given that the widely accepted duration of action of 0·3 mg/kg methadone is 3 to 4 hours and the rapid elimination of methadone in dogs (1·75 to 4·3 hour half life, Kukanich and Borum 2008), this was not considered a confounding factor.

Dogs recruited into this study demonstrated unilateral thoracic limb lameness. Although the incidence of bilateral disease with elbow dysplasia is high (Fitzpatrick and Yeadon 2009), not all affected dogs demonstrate bilateral lameness. Examination of PVFi in this population of dogs showed a clear difference between the lame limb (7·72 ±1·76 N/kg) and the contralateral (10·57 ±1·39 N/kg) limb, which are consistent with values obtained by Hercock and others (2009).

When the relationship between LNRS and PVFi was examined in our study, non-lame dogs were shown to have a PVFi>7 N/kg. When these data are examined as individual value plots, the variation in PVFi is wide and PVFi does not correlate with LNRS, which is in agreement with previous work (Quinn and others 2007, Waxman and others 2008).

Significant differences in ground reaction forces exist between Labrador retrievers and Rottweilers which are attributable to differences in conformation and body weight (Molsa and others 2010). In the present study data were collected from a variety of breeds of different conformations and weights. Our sample size calculation was based on PVFi data from lame and sound Labradors and may explain why no significant difference was detected between groups and represents a limitation of this clinical study. In this clinical study it would have been unrealistic to limit recruitment to one breed.

The use of meloxicam and methadone in this anaesthetic protocol may have affected the ability to detect pain and lameness; however, the inclusion of a placebo group, or withholding of a non-steroidal anti-inflammatory in this clinical study, was considered unacceptable in these client-owned dogs.

In conclusion, in this study with group sizes similar to previous studies documenting a positive effect of IA agents, it was not possible to demonstrate a significant benefit to either IA treatments using PVFi as an outcome measure. Changes in pain scores over time demonstrate a beneficial effect of bupivacaine at both time points and morphine at the 24-hour time point. Data reported for ST, ROL and ROUL should be incorporated with PVFi results for sample size calculations for future studies.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. STATISTICAL METHODS
  6. Results
  7. DISCUSSION
  8. Acknowledgements
  9. Conflict of interest
  10. References

The authors are grateful to Petsavers for supporting this study with a Petsavers Clinical Research Project Grant and to Mr Carl Bradbrook for reviewing the videos for lameness scoring. The study was conducted at the University of Liverpool.

Conflict of interest

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. STATISTICAL METHODS
  6. Results
  7. DISCUSSION
  8. Acknowledgements
  9. Conflict of interest
  10. References

None of the authors of this article has a financial or personal relationship with other people or organisations that could inappropriately influence or bias the content of the paper.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. STATISTICAL METHODS
  6. Results
  7. DISCUSSION
  8. Acknowledgements
  9. Conflict of interest
  10. References