Comparison of surgical and non-surgical treatments for 3- and 4-part proximal humerus fractures: A network meta-analysis

Abstract

Introduction

Common treatment strategies for proximal humerus fractures include non-surgical treatment, open reduction internal fixation, hemiarthroplasty, and reverse total shoulder arthroplasty. There is currently no consensus regarding the superiority of any one surgical strategy. We used network meta-analysis of randomized controlled trials to determine the most successful treatment for proximal humerus fractures.

Methods

MEDLINE, EMBASE, Web of Science, and Cochrane Central electronic databases were searched for randomized controlled trials comparing 3- and 4-part proximal humerus fracture treatments. Data extraction included the mean and standard deviation of clinical outcomes (Constant, DASH), adverse events, and additional surgery rates. Standard Mean Difference was used to compare clinical outcome scores, and pooled risk ratios were used to compare adverse events and additional surgeries.

Results

Eight randomized controlled trials were included for network meta-analysis. Non-surgical treatment was associated with a lower rate of additional surgery and adverse events compared to open reduction internal fixation. Reverse total shoulder arthroplasty resulted in fewer adverse events and a better clinical outcome score than hemiarthroplasty. Non-surgical treatment produced similar clinical scores, adverse event rates, and additional surgery rates to hemiarthroplasty and reverse total shoulder arthroplasty.

Conclusion

Non-surgical treatment results in fewer complications and additional surgeries compared to open reduction internal fixation. Preliminary data supports reverse total shoulder arthroplasty over hemiarthroplasty, but more evidence is needed to strengthen this conclusion.

Objectives

The incidence of proximal humerus fractures (PHFs) has been increasing since the 1970s.^1,2 One-fifth of PHFs are categorized as 3-part and 4-part fractures – the most severe and difficult to treat according to Neer's four-segment classification system.^3,4 Current treatment strategies for 3- and 4-part fractures include non-surgical treatment (NST), open reduction internal fixation (ORIF), hemiarthroplastry (HA), and reverse total shoulder arthroplasty (rTSA).⁵ Typical NSTs involve bracing the shoulder with a standard sling, cuff and collar, and occasionally a hanging arm cast or airplane splint.⁶ ORIF encompasses a wide array of techniques, from intramedullary nailing to osteosynthesis with a locking plate. This is the preferred method of treatment especially for younger patients; however, it has been linked to higher rates of non-union, mal-union, and loss of reduction with hardware impingement in osteoporotic bone.^7,8 HA is another treatment option that may have some benefit if the proximal humerus fragments are deemed non-reconstructable.⁹ While pain relief with HA is generally good, patients can suffer from rotator cuff deficiency or limited motion if the humeral tuberosities do not heal well.⁹ Lastly, rTSA involves the use of both humeral head and socket prostheses with a reversal of the normal ball and socket orientation. The rTSA is becoming the treatment of choice for patients with non-reconstructable PHFs as patient can achieve excellent motion, even when the tuberosities do not heal or if the rotator cuff is incompetent. It may also be performed for NST fracture sequelae and the revision of failed HA procedures.¹⁰

While evidence that favors NST over surgical intervention is mounting, the superiority of any one surgical strategy has yet to be determined.¹¹ Additional research and subsequent formulation of guidelines is necessary to optimize patient care for patients with PHFs. Randomized controlled trials (RCTs) have long been thought of as the “gold standard” of medical research.¹² With limited RCTs available for the treatment of 3-part and 4-part PHFs, previous network meta-analyses have resorted to including non-RCT studies as well as using wide age ranges,¹³ which may have diluted the reliability of their findings.

The purpose of this network meta-analysis (NMA) was to compare RCTs of various treatments of 3-part and 4-part PHFs in patients over 51 years old. We aim to determine the most successful treatment modality based on functional outcome scores as well as adverse event and additional surgery rates.

Materials and methods

Data sources

In accordance with the Cochrane Collaboration and PRISMA guidelines, in September 2016, we searched the MEDLINE, EMBASE, Web of Science, and Cochrane Central electronic databases for all relevant studies regardless of publication date. The following search strategy was created for MEDLINE and adapted for other databases: (exp Shoulder Fractures/ OR ((fracture$ ADJ3 proximal) AND humer$).ti,ab. OR (shoulder ADJ3 fracture$).ti,ab.) AND (exp Shoulder Fractures/su OR exp Fracture Fixation, Internal/ OR exp Arthroplasty/ OR (surg$ OR operative OR fixation OR arthroplast$ OR hemiarthroplast$ OR osteosynthesis OR shoulder prothes$ OR replace$ OR orif OR rsa OR rtsa).ti,ab) AND exp Aged/ OR ((old$ OR aged) ADJ1 (patient$ OR individual$ OR adult$ OR men OR women OR person$ OR people)).ti,ab OR (elder$ OR geriatric$ OR age 65 OR age 70 OR age 80 OR age 85 OR age 90 OR over 65 OR over 70 OR over 75 OR over 80 OR over 85 OR over 90 OR late life).ti,ab.

The search was updated in April 2018 and no new RCTs were eligible for analysis.

Study selection

We included studies that met the following inclusion criteria: (1) RCTs, (2) available in English language, (3) 3-part or 4-part PHFs, (4) using one of the following for management: NST, ORIF with locking plate, HA, and rTSA, and (5) minimum follow-up time of 12 months. The primary outcome of interest was functional outcome scores, with adverse event rates and additional surgery rates as secondary outcomes. The functional outcome scores used were Constant and DASH scores, and these were compared using standardized mean difference (SMD), a method that allows comparison of different scoring systems (e.g. Constant and DASH) against each other.

Critical appraisal

Two independent reviewers evaluated the methodological quality of the included studies using the Jadad scale, a tool developed for assessing the quality of RCT by assigning a score between 0 and 5 based on level of blinding.¹⁴ Data synthesis was only done on higher quality studies (defined as a Jadad score ≥2) and entered into meta-analysis. Any disagreement between investigators about study selection or evaluation was resolved by discussion and consensus.

Data extraction

Data collection was based on the Cochrane Consumers and Communication Review Group's data extraction template. The mean and standard deviation of clinical outcome assessments including Constant–Murley Shoulder Outcome score,¹⁵ and the Disabilities of the Arm, Shoulder and Hand (DASH) score,¹⁶ as well as the rate of adverse events and additional surgeries were synthesized. We did not attempt to obtain raw data or confirm synthesized data from investigators of the included studies.

The following adverse events were pooled in this study: avascular necrosis of humeral head, hematoma or infection, nonunion, secondary displacement, implant failure, primary or secondary screw penetration, proximal migration of humeral head, and greater tuberosity mal-union or resorption. The pooled additional surgery rate was reported to avoid double counting for cases that underwent additional surgery to treat an adverse event. The rate of adverse events and additional surgeries were calculated using the number of patients receiving the treatment.

Data synthesis

Statistical analysis was performed using Comprehensive Meta-Analysis Version 2 (Biostat, Englewood, NJ, USA). To assess heterogeneity among individual included studies, we calculated Cochran's Q statistic. Due to concerns about low sensitivity of Cochran's Q for assessing heterogeneity, a P-value of 0.10 or less was set to determine statistical significance.¹⁷ The I² statistic was also reported and classified to represent low (0%–25%), moderate (25%–50%), substantial (50%–75%), or considerable (>75%) inconsistency.¹⁸ In the case of significant heterogeneity, a random effect analysis was employed. Otherwise a fixed effect analysis was employed. A funnel plot was drawn and inspected visually to assess publication bias and Egger's test (with a P-value of 0.05 considered statistically significant) was used to evaluate for the presence of publication bias.

We reported pooled complication rates for additional surgeries and adverse events with 95% confidence interval (CI). The number needed to harm (NNH) is reported when the P-value of the pooled risk ratio (RR) is < 0.05. The NNH indicates the number of patients harmed when exposed to a certain intervention, beyond what a control treatment would cause.

Data are represented via network plots. Each circle represents an intervention. The size of the circle roughly corresponds to the number of pooled participants for that intervention. Each line represents a direct treatment comparison made via RCT(s). Bolded lines indicate significant findings, and arrows point to the treatment modalities that exhibit significantly higher Constant/DASH scores, adverse event rates, or additional surgery rates. Dashed lines indicate no significant difference between interventions.

Results

Search results

A total of 3617 potentially relevant publications were identified by searching the MEDLINE, EMBASE, Web of Science, and Cochrane Central databases (Figure 1). An initial screening allowed for exclusion of 3574 studies based on title and abstract, while the remaining 43 studies were retrieved for detailed assessment. An additional two publications were identified from contacting authors. A total of 37 articles did not meet the inclusion criteria and were ultimately excluded (Figure 1).

Figure 1.

Study selection flow chart. A flow chart describing the process by which studies were identified, screened, and included for meta-analysis.

Study characteristics

Eight RCTs including a total of 364 patients with 364 fractures were included for NMA (Table 1). The mean age was 73.4 years, with a minimum age of 51 years in the six studies that reported an age range. All authors except Zyto et al. confirmed fracture classification using experienced orthopedic surgeons, experienced radiologists, and CT scans. It should be noted that there were two distinct 2011 studies published by Olerud et al., one comparing HA to NST and another comparing ORIF to NST, which will be referred to as Olerud et al.(a)¹⁹ and Olerud et al.(b),²⁰ respectively. Five studies^19–23 used NST with a total of 108 patients, five studies^{19,21,24–26} used HA with a total of 118 patients, five studies^20,22–25 used ORIF with a total of 107 patients, and one study²⁶ used rTSA with a total of 31 patients.

Table 1.

Studies included for meta-analysis.

Study	Country	Mean age (range)	Study size	Control (N)	Treatment (N)	Fracture	Follow-up timea	Outcome measures
Boons et al.21	Netherlands	78.2 (N/A)	47	NST (23)	HA (24)	4-part	12 months	Constant, adverse events, reoperation
Cai et al.24	China	71.9 (67–86)	27	ORIF (15)	HA (12)	4-part	24 months	Constant, DASH, adverse events, reoperation
Chen et al.25	China	66 (51–81)	56	ORIF (28)	HA (28)	4-part	24–48 months	Constant, DASH, adverse events, reoperation
Fjalestad and Hole22	Norway	72.6 (60–88)	42	NST (19)	ORIF (23)	3- and 4-part	24 months	Constant, adverse events, reoperation
Olerud et al.(a)19	Sweden	77 (58–92)	49	NST (25)	HA (24)	4-part	24 months	Constant, DASH, adverse events, reoperation
Olerud et al.(b)20	Sweden	74 (56–92)	53	NST (26)	ORIF (27)	3-part	24 months	Constant, DASH, adverse events, reoperation
Sebastia-Forcada26	Spain	74 (70–85)	61	HA (30)	rTSA (31)	3- and 4-part	24 months	Constant, DASH, reoperation
Zyto et al.23	Sweden	74 (N/A)	40	NST (15)	ORIF (14)	3- and 4-part	36 months	Constant

^aOutcomes at latest follow-up dates available were used for meta-analysis; N/A: none available; NST: non-surgical treatment; ORIF: open reduction internal fixation; HA: hemi-arthroplasty; rTSA: reverse total shoulder arthroplasty.

There was no significant heterogeneity among included studies and the probability of publication bias was not significant per Egger's test (p = 0.21). The Jadad scoring and critical appraisal results are shown in Tables 2 and 3, respectively. The average Jadad score was 3.1, indicating the included studies were largely of high quality.¹⁴ According to the Cochrane Collaboration's Tool for Assessing Risk of Bias,¹¹ all included studies except for Cai et al. showed no risk of selection bias. However, all studies except for Sebastia-Forcada et al. were found to be at risk for performance and detection bias due to a lack of participant and outcome blinding, respectively. Additional sources of bias were self-reported in all eight studies (Table 3).

Table 2.

Evaluation of study quality using Jadad scale.

Study	Randomization			Blinding		Account of all patients	Total score
	Randomization mentioned	Appropriate method of randomization	Inappropriate method of randomization	Blinding mentioned	Appropriate method of blinding	Fate of all patients in the trial is known
Boons et al.21	Yes	Yes	No	No	No	Yes	3
Cai et al.24	Yes	No	No	No	No	Yes	2
Chen et al.25	Yes	Yes	No	No	No	Yes	3
Fjalestad and Hole22	Yes	Yes	No	No	No	Yes	3
Olerud et al.(a)19	Yes	Yes	No	No	No	Yes	3
Olerud et al.(b)20	Yes	Yes	No	No	No	Yes	3
Sebastia-Forcada26	Yes	Yes	No	Yes	Yes	Yes	5
Zyto et al.23	Yes	Yes	No	No	No	Yes	3

Table 3.

Critical appraisal of included studies according to the Cochrane Collaboration's tool for assessing risk of bias.

	Boons et al.21	Cai et al.24	Chen et al.25	Fjalestad and Hole22	Olerud et al.(a)19	Olerud et al.(b) 20	Sebastia- Forcada26	Zyto et al.23
Selection bias
Random sequence generation	+	−	+	+	+	+	+	+
Allocation concealment	+	−	+	+	+	+	+	+
Performance bias
Blinding of participants and personnel	−	−	−	−	−	−	+	−
Detection bias
Blinding of outcome assessment	−	−	−	−	−	−	+	−
Attrition bias
Incomplete outcome data	+	−	+	+	−	−	+	−
Reporting bias
Selective reporting	−	−	−	−	−	−	−	−
Other bias
Other sources of bias	+	+	+	+	+	+	+	+

Constant/DASH outcome

The SMDs comparing Constant/DASH scores at the furthest available follow-up time are shown in Figure 2. Based on one study²⁶ with 61 patients, rTSA resulted in significantly better outcomes (i.e. better Constant/DASH scores) compared to HA (SMD = 0.89; CI = 0.36–1.41; p < 0.01), but there were no significant differences between HA and ORIF. Similarly, ORIF and HA did not produce significantly better outcomes than NST, based on three studies with 124 patients and two studies^19,21 with 96 patients, respectively. Using NMA, additional indirect comparison between HA and ORIF showed no significant difference in Constant/DASH score (p = 0.52).

Figure 2.

Constant/DASH score standardized mean difference (95% CI). Bold lines indicate significant findings. Dashed lines indicate insignificant findings. Arrows point to the treatment modality that exhibits significantly higher Constant/DASH scores.

NST: non-surgical treatment; ORIF: open reduction internal fixation.

Adverse event rates

The pooled RRs comparing adverse event rates are shown in Figure 3. Given that every patient who received treatment was potentially at risk for an adverse event, all study participants were included for the purposes of calculating adverse event and additional surgery rates, irrespective of whether they were present for follow-up. Based on one study²⁶ with 61 patients, rTSA was associated with a significantly lower adverse event rate than HA (RR = 0.57; CI = 0.36-0.90; NNH = 3.1; p = 0.02). However, based on three studies^20,22,23 with 150 patients, ORIF was associated with a significantly higher adverse event rate than NST (RR = 1.45; CI = 1.10–1.91; NNH = 3.5; p < 0.01).

Figure 3.

Adverse event rate pooled risk ratio (95% CI). Bold lines indicate significant findings. Dashed lines indicate insignificant findings. Arrows point to the treatment modality that exhibits a higher rate of adverse events.

NST: non-surgical treatment; ORIF: open reduction internal fixation.

There were no significant differences between HA versus NST and HA versus ORIF. Additional indirect comparison between HA and ORIF showed no significant difference in adverse event rates (p = 0.64).

Additional surgery rates

The pooled RRs comparing additional surgery rates are shown in Figure 4. Based on three studies^20,22,23 with 150 patients, ORIF was associated with a significantly higher rate of additional surgery compared to NST (RR = 8.13; CI = 2.10–31.60; p < 0.01). There were no significant differences between rTSA versus HA, HA versus NST, and HA versus ORIF. Additional indirect comparison of HA and ORIF showed no significant difference in additional surgery rate (p = 0.21).

Figure 4.

Additional surgery rate pooled risk ratio (95% CI). Bold lines indicate significant findings. Dashed lines indicate insignificant findings. Arrows point to the treatment modality that exhibits a higher rate of additional surgery.

NST: non-surgical treatment; ORIF: open reduction internal fixation.

Conclusions

The most important finding of our study was that NST was associated with a lower rate of adverse events (RR = 1.45; NNH=3.5; p < 0.01) and additional surgery (RR = 8.13; p < 0.01) compared to ORIF, and in all other respects it performed similarly to the surgical interventions (Table 4). In other words, for every four patients undergoing ORIF, one will encounter an adverse event beyond NST. Another important finding in our study was that among the surgical interventions, rTSA produced better functional outcomes than HA (SMD = 0.89; p < 0.01), a lower rate of adverse events (RR = 0.57; p = 0.02), and no difference in rate of additional surgery (RR = 0.17; p = 0.09) (Table 4). Besides the aforementioned findings, there were no significant differences in Constant/DASH scores, adverse events, or reoperations between NST, ORIF, HA, and rTSA (Table 4).

Table 4.

Network meta-analysis statistical results for intervention comparisons.

	Constant/DASH score		Additional surgery		Adverse events
	SMD (95% CI)	P-value	RR (95% CI)	P-value	RR (95% CI)	P-value
ORIF vs. NST	0.04 (−0.31 to 0.41)	0.81	1.45 (1.10 to 1.91)	<0.01	8.13 (2.1 to 31.60)	<0.01
HA vs. NST	0.22 (−0.18 to 0.62)	0.28	1.32 (0.69 to 2.50)	0.40	1.98 (0.36 to 10.97)	0.43
ORIF vs. HA	0.26 (−0.18 to 0.70)	0.25	0.96 (0.35 to 2.63)	0.94	1.40 (0.40 to 4.53)	0.58
HA vs. rTSA	0.89 (0.36 to 1.41)	<0.01	0.57 (0.36 to 0.90)	0.02	0.17 (0.21 to 1.81)	0.09

NST: non-surgical treatment; ORIF: open reduction internal fixation; HA: hemi-arthroplasty; rTSA: reverse total shoulder arthroplasty; SMD: standardized mean difference; RR: risk ratio; CI: confidence interval.

Displaced PHFs are common occurrences among the aging population. Despite the lack of consensus regarding treatment,^27–30 the incidence of surgical interventions has almost quadrupled in the last 20 years.³¹ It remains challenging to identify which patient would benefit from NST. A 2015 Cochrane review by Handoll and Brorson compared treatment strategies for PHFs.²⁷ Handoll and Brorson found high-to-moderate quality evidence that surgery does not produce better outcomes than NST.²⁷ Unlike our study, they pooled data from randomized and quasi-RCTs. An additional difference was that they included studies that compared different surgical techniques for the same type of procedure (e.g. deltoid-split approach versus deltopectoral approach for ORIF) as well as studies that compared different surgical equipment for the same type of procedure (e.g. polyaxial versus monoaxial screws for ORIF).²⁷ We only analyzed direct head-to-head comparisons of different treatment types. Consequently, our analysis included fewer studies and a smaller cohort of patients, but our narrower inclusion criteria should provide a greater degree of reliability to our findings.

Handoll and Brorson²⁷ reported insufficient evidence to support any single surgical strategy. Based on our results, however, rTSA should be preferred to HA, since it was associated with a significantly lower adverse event rate (RR = 0.57; p = 0.02). This is in agreement with a recent NMA by Chen et al. that also found rTSA produced favorable outcomes compared to HA, ORIF, and NST.¹³ It should be noted that Chen et al. included non-RCT studies, 2-part PHFs, and patients as young as 16 years old. Our analysis improved upon these methodological limitations. First, we attempted to capture the highest quality data by exclusively analyzing RCTs. Second, 2-part PHFs were excluded from our analysis. Most 2-part PHFs can be treated with closed interventions unlike 3- and 4-part fractures, which are more complex and therefore may require separate grouping for analysis.³ Third, the minimum age of our patient population from studies that reported ranges was 51, rather than 16, which should better represent the population most affected by PHFs. A recent epidemiological study showed that the incidence of PHFs is only 13 per 100,000 for women between the ages of 20–29, increasing to 112 for the 50–59 age group and 379 for the 80 + age group.³² Outcomes for patients between 50 and 65 years old may be different than outcomes for patients older than 65; however, there is not enough data to analyze these demographics separately. Therefore, our analysis ended up including patients above 51 with a mean age of 73.4 to capture the increased incidence of PHFs in the elderly as accurately as possible. Age is an important consideration because rTSA is usually reserved for patients aged 70 or older;¹⁰ therefore, Chen et al.'s inclusion of younger patients may have resulted in unbalanced comparisons between treatments, something our analysis improves upon.

Several studies have demonstrated the lack of consensus among surgeons regarding treatment modalities for PHFs. There is regional variation,²⁸ variation between surgeons at the same institution,²⁹ and even variation in a single surgeon's decision making depending on the number of treatment options available.³⁰ Our study adds to the accumulating body of evidence favoring NST. The recently published PROFHER trial, which we were unable to include in this study due to their use of 1- and 2-part PHFs, also found that there were no differences in clinical outcomes between surgical and non-surgical intervention.³³ NST has been shown to be less costly and to have a higher probability of cost effectiveness compared to surgery,³⁴ with our study and the PROFHER trial showing similar outcomes between the two interventions, there is an argument to be made in favor of NST. In the event that NST is not indicated, our study supports previous findings in favor of rTSA as a primary surgical intervention.¹⁰ It should be noted that rTSA tends to be the more expensive option,³⁵ but its cost effectiveness is still superior to that of HA.³⁶

This NMA has several limitations. First, rTSA was only represented in one head-to-head study, while ORIF (5), HA (5), and NST (5) were more equally represented. Thus, an important finding, namely that rTSA resulted in better Constant/DASH and adverse event rates than HA, was predicated on the single study by Sebastia-Forcada et al.²⁶ Second, one²⁴ of the eight included studies received a low Jadad score due to issues with blinding and randomization (Table 2). Nevertheless, this study was still included for analysis due to the overall low number of available RCTs. Third, the standard deviation was not provided in all studies so we had to input these values based on the Cochrane Collaboration's guidelines.¹¹ Fourth, rTSA has been shown to have a significant learning curve with inexperienced surgeons having a higher rate of complications than those with more experience.³⁷ As most of the studies included in this paper were performed in academic medical centers, it is possible that the advantages of rTSA over HA may have been influenced by surgeon experience. Fifth, functional outcomes may be largely influenced by patient age and comorbidity. Because individualized patient data were not available from the included RCTs, we were unable to control for differences in these factors. Sixth, we did not include osteosynthesis with a proximal humeral nail in our analysis. Future studies should incorporate this treatment modality into their analysis, as at least one RCT has found intramedullary nailing to produce similar clinical outcomes to treatment with locking plates for 2- and 3-part PHFs.³⁸ Lastly, the included studies had follow-up times ranging from 12 months to 48 months. These follow-up times, while not inadequate, may fail to capture later adverse events and reoperations associated with the limited lifespan of some implants.

In conclusion, our analysis found that NST produced lower adverse event rates than ORIF. Furthermore, rTSA produced encouraging outcomes compared to HA, but these results were predicated on one study. Currently, there are two RCTs under way that should yield more data. The PROFHER-2 trial is aiming to recruit 380 patients by 2021 to compare HA versus rTSA versus NST.³⁹ Published results can be expected by 2023. Another trial is already underway comparing HA to rTSA in 50 patients older than 65.⁴⁰ This trial ends in 2019. We recommend that future RCTs not only compare rTSA to HA and NST but also directly to ORIF in the hope that thorough evidence-based treatment guidelines can be developed.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.