2.1 Search procedure

A comprehensive review of major scientific databases was conducted in search of studies evaluating core decompression alone or core decompression with BMAC for avascular necrosis of the femoral head. On November 20, 2023, PubMed, Web of Science, Embase, Scopus, and Cochrane were searched for all available articles with no limitations on the date of publication.

MeSH was utilized to identify key terms, which were subsequently applied to Boolean operators. The search string utilized was: (“Core decompression”) AND (“AVN” OR “ANFH” OR “avascular necrosis” OR “osteonecrosis” OR “Femoral necrosis”) AND (“Clinical Outcomes” OR “Harris Hip Score” OR “HHS” OR “Western Ontario and McMaster Universities Arthritis Index” OR “WOMAC” OR “Copenhagen Hip and Groin Outcome Score” OR “HAGOS” OR “Visual Analogue Scale” OR “VAS” OR “iHOT-12” OR “iHOT-33” OR “international hip outcome tool”) AND (“BMAC” OR “Bone marrow aspirate” OR “mesenchymal stem cells” OR “Autogenous bone marrow aspirate” OR “bone marrow concentrate” OR “bone marrow stromal cells”).

In addition, Google Scholar was manually searched for gray literature which yielded no novel literature. This search yielded 855 records which were imported into Rayyan. ai to detect and remove duplicates using the 99% match “Detect Duplicates” function. There were 519 duplicates identified and removed, resulting in 336 remaining studies for screening of inclusion and exclusion criteria. Titles and abstracts were independently screened by two authors (HB and SS), which led to the exclusion of 303 articles. The remaining 33 articles were subjected to full-text appraisal by the same two reviewers.

2.2 Inclusion and exclusion criteria

Articles selected for inclusion in the analysis were randomized control trials, retrospective reviews, and comparative studies that evaluated CD or CD with BMAC for treatment of ANFH. Studies that examined BMAC or autologous mesenchymal stem cell (MSC) implantation were included for data extraction for the BMAC group. While BMAC contains growth factors not present in MSC, the primary therapeutic component is the same. Literature often uses these phrases interchangeably, and thus both paper types were included for completeness. Articles needed to record both pre- and post-intervention HHS or WOMAC scores to determine the effects of the intervention on overall physical functionality, pain severity, absence of deformity, range of motion, and stiffness. Excluded studies were those that did not have full-text availability, those that did not have full data availability, and those that did not have an English translation. Furthermore, case reports, case series, systematic reviews, and review articles were excluded. Of the remaining studies, 33 were reviewed in their entirety before elimination due to lack of control group (n = 8), combined or additional interventions (n = 8), wrong outcome reported (n = 6), and no useable data (n = 1) (Fig. 1). For the remaining 9 studies, HB and SS extracted the relevant data to be used for statistical analysis.

Close

Fig. 1 PRISMA 2020 flow chart detailing study collection process.

Export to PPT

2.3 Data collection and analysis

Qualitative and quantitative data collection began once study selection was completed. HHS and WOMAC were selected as the primary variables for our study. HHS is a 100-point scale where a higher score represents better functional outcomes, which is the opposite of WOMAC. Therefore, WOMAC scores were reversed in the pre-post statistical analysis so that pooled effect size could be measured correctly as a positive or negative shift. HHS quantifies the functional outcomes of hip pain, function, absence of deformity, and mobility where 91 of 100 possible points are derived from patient pain and function while 4 and 5 points are derived from absence of deformity, and mobility respectively. WOMAC also favors pain and function in its score in that it assesses pain out of 20 points, physical function out of 68 points, and stiffness out of 8 points. Both outcome variables were included for completeness.

Mean, standard deviation, and sample size of the HHS and WOMAC data were collected at baseline and at 4 ± 2.5 years follow-up. Statistical analysis was performed using a meta-analysis with a random effects model using IBM SPSS Statistics for Windows, version 29 (IBM Corp., Armonk, N.Y., USA). This approach pooled the effect sizes of individual studies, allowing for evaluation of the mean changes in HHS and WOMAC scores relative to variation. A p-value of less than 0.05 was utilized to determine statistical significance. The extent of improvement in HHS or WOMAC scores was represented by the pooled and individual effect size (Cohen's d) with 95% confidence intervals (95% CI [LL,UL]). Cohen's d was used to represent the effect size, as seen in the forest plot (Fig. 2, Table 1).

Close

Fig. 2 Forest plot of the random effects meta-analysis showing Group 1 (core decompression alone) compared to Group 2 (core decompression with BMAC).

Export to PPT

Heterogeneity of the study results, representing variance, was primarily assessed using Q-statistics and the I2 ratio (I2 = τ2/H2) (Fig. 2). A test of subgroup homogeneity was used to represent differences between the control (CD) and intervention (CD with BMAC) groups, where statistical significance signifies that a difference is present. A larger I2 suggests more variance between study results, with observed differences potentially coming from another variable, such as bias in the study design. Tau-squared (τ2) represents the absolute variation between effect sizes, without considering the variation that is expected from random chance. A random effects model was utilized; this allows for analysis of variance greater than what would be expected from chance, thus delineating the influence of an external variable different from the influence of the dependent variable, whereas the less conservative common effects model does not. An H2 ratio was also reported to analyze variance, where a value of 1 represents equivalent variance between the fixed and random effects models. If this were the case, this would represent low heterogeneity, whereas there is no variation in study effect sizes greater than what is expected from random chance alone.

2.4 Risk of bias and certainty of evidence assessment

Included manuscripts were evaluated for methodological quality using modified Grading of Recommendations Assessment, Development and Evaluation (GRADE) criteria.¹³ Bias in the included articles was assessed independently by two authors (HB and SS) based on their respective study design. Since randomized control trials, retrospective reviews, and cohort studies were included in our analysis, they were subjected to evaluation with ROBINS-I.¹⁴ The data is presented in plot format (Figs. 3 and 4).

Close

Fig. 3 ROBINS-I risk of bias traffic light plot.¹⁵.

Export to PPT

Close

Fig. 4 Summary plot of the ROBINS-I risk of bias assessment.¹⁵.

Export to PPT

3.1 Measures of effect

Each of the included studies reported a significant improvement in functional outcome scores, except for Hauzeur 2019. The CD group reported significant (p = 0.01) and large (Cohen's d = 3.80 [0.83, 6.76]) improvements in functional outcome scores between baseline and follow-up. The CD with BMAC group also reported significant (p < 0.01) and large (Cohen's d = 3.34 [1.02,5.66]) improvements in functional outcome scores between baseline and follow-up. Difference between effect sizes was Cohen's d = 0.44, but subgroup analysis found no significant difference between groups (p = 0.81) (Fig. 2, Table 1).

3.2 Heterogeneity

In the CD only group, τ2 = 15.6, H2 = 111.6, and I2 = 99.6%; the CD with BMAC group reported τ2 = 5.5, H2 = 58.1, and I2 = 98.3%. The design variables, which may have affected the variance in results between studies, were examined in the risk of bias and quality of evidence assessments.

3.3 Risk of bias and certainty of evidence assessment

Modified GRADE analysis revealed a very low quality of evidence for two studies, a low quality of evidence for five studies, a moderate quality of evidence for one study, and a high quality of evidence for one study. The non-randomized studies (n = 7) started off with an initial low rating due to inherent risk of bias.

Assessment of bias in each study was performed using Cochrane's ROB-1 tool, which revealed a low risk of bias in five studies and a moderate risk of bias in two studies (Figs. 3 and 4). The determination of risk of bias was made by two independent reviewers according to seven domains: bias due to confounding variables, bias due to selection of participants, bias due to classification of interventions, bias due to deviations from intended interventions, bias due to missing data, bias in measurement of outcomes, and bias in selection of the reported results. Two studies (Talmac 2018 and Zhao 2015) were found to have some concerns due to confounding variables. Two studies (Hauzeur 2019 and Tabatabaee 2015) were found to have some concerns due to bias in the measurement of outcomes. Since overall risk of bias for a study is determined based on the lowest rating in each domain, these studies were determined to have a moderate risk of bias. The other five studies were found to have a low risk of bias across all domains.

4.1 Limitations

A major limitation to the study is lack of standardization of practices. Without standardization of aspirate volumes, exact procedure, and cell line quality in the aspirate, it is difficult to establish a strong comparison of results. This lack of standardization along with small sample sizes makes it more difficult to generalize results. Large, high quality RCTs would help to correct this issue. Without data on patients' medical history and home medications, confounding factors such as long-term oral glucocorticoids, chronic pain medications, nutritional deficiencies, and other medical conditions not disclosed in the reviews may present as confounding factors. Finally, while the usage of imaging modalities such as MRI allows researchers to visualize the joint, the subjective nature of pain and a patient's individual ability to endure likely has effects on functional status.