Early postoperative laboratory parameters are predictive of initial treatment failure in acute septic arthritis of the knee and shoulder joint

Univariate analysis

249 patients with septic monoarthritis, including 194 knee (77.9%) and 55 shoulder joints (22.1%) were analyzed in total. Sixty-five (26.1%) experienced failure of a single surgical debridement (25.3% of shoulders, 29.1% of knees). Median follow-up for the single-surgery group was 34 months, and 21 for the failed single-surgery group, respectively. Median time to second surgery was 15 days. Infectious etiology could be retraced in 141 patients, including 61 iatrogenic (56.5%), 46 hematogenic (42.6%), and one traumatic case (1.2%). There were no significant differences regarding gender, age, BMI, fever at admission, Gächter score, and surgery duration between successful and failed single surgical debridement (p > 0.05). Higher CCI scores were significantly associated with the need for repeated surgery (p = 0.025). Further, longer symptom duration (p = 0.003) and severe osteoarthritis, as reflected by high grades of the Kellgren-Lawrence classification (cf. Figure 2), were significantly associated with failure of a single debridement (p = 0.013). Table 1 summarizes the results in detail.

Figure 2

Distribution of Kellgren-Lawrence classes in the success and failure group.

Of the 65 patients experiencing treatment failure, 39 (60%) were successfully reoperated, 18 (27.7%) required further surgical interventions for septic arthritis, 4 (6.2%) presented with impaired wound healing, 3 (4.6%) died in the postoperative course, 2 (3.1%) were admitted to the ICU for septic shock. One (1.5%) knee arthrodesis was performed.

Surgical technique

In total 89 arthrotomies (35.7%; 35 shoulders, 54 knees) and 160 arthroscopies (64.3%; 20 shoulders, 140 knees) were performed. The surgical technique had a significant impact on shoulder joints, with failure rates of 50.0% for arthroscopy compared to 17.1% for arthrotomy (p = 0.01). The comparison of baseline characteristics between both treatment strategies showed a significantly higher percentage of female patients (50.6% vs. 36.2%, p = 0.04), a higher mean age (67.0 vs. 59.3 years, p = 0.003), as well as a higher CCI in the arthrotomy group (4.28 vs. 3.09, p < 0.001). Arthroscopy was performed more frequently in knee joints (60.7 vs. 87.5%, p < 0.001), and average Gächter scores were higher in the arthrotomy group (p = 0.011).

Microbiological analysis

Positive cultures were found in 37.9% in the successful single debridement group, and in 66.15% in the failure group respectively (p < 0.01). In patients in whom no pathogens could be detected, the diagnosis was confirmed by intraoperative histopathological examination. Staphylococcus aureus infections showed no significant influence on treatment failure (p = 0.187). Methicillin-resistant Staphylococcus aureus (MRSA) infections were not separately analyzed, as there were too few cases (cf. Figure 3).

Figure 3

Distribution of pathogens over a total of 112 positive bacterial cultures.

Postoperative laboratory parameters

CRP values on the third and fifth postoperative day were significantly higher in the treatment failure group. Neutrophilic granulocytes three days after surgery, and hemoglobin levels before discharge showed a significant difference in absolute as well as relative measures (cf. Table 3 and Fig. 4).

Table 3 Laboratory values described by mean (SD) and compared between successful treatment and failure groups.Figure 4

Absolute (abs) and relative (rel) CRP changes compared to admission values.

Score development

Two different SYNC (Symptom duration, Neutrophilic granulocytes, Charlson comorbidity index, 3rd day CRP, 5th day CRP) scoring systems were developed: the SYNC3 includes influencing factors that are known as early as day 3 after surgery, the SYNC5 is a modified version additionally including CRP drop between admission and the fifth postoperative day. Calculated weights for each of the variables are given in Table 2. The first score reached an AUC of 0.80 (p < 0.001, CI = [0.69 0.91]), the second had an AUC of 0.845 (p < 0.001, CI = [0.72 0.97]) , and both can be considered as excellent in terms of discrimination. Corresponding ROCs are shown in Fig. 5. Positive correlation of risk score categories and failure rates was underlined by the Cochran-Armitage trend test with p < 0.0001 for both scores. Patients were grouped into low, intermediate, and high risk of failure based on their score results (cf. Figure 6). Additionally, a website for score calculation was created (www.ortho-score.com).

Figure 5

ROCs from the two prediction models. The first includes information on CRP and neutrophilic granulocytes up to day 3 post surgery, symptom duration, CCI, and bacterial culture result. The second model additionally includes information on CRP change from admission to day 5 post-OP.

Figure 6

Low, intermediate, and high risk of failure depending on score points on the 3rd and 5th postoperative day.

Discussion

Septic arthritis of large joints is potentially life-threatening, and requires repeated intervention if a single irrigation and debridement was not sufficient to control the infection to prevent further joint destruction19. To the best of our knowledge, we evaluated the largest patient cohort on risk factors for failure of a single arthroscopy or arthrotomy for septic arthritis to date19,20,21,23. A new scoring system for treatment failure risk estimation was created based on symptom duration, CCI, bacterial culture results, postoperative change in CRP, and change in neutrophilic granulocytes.

This is the first study to demonstrate that higher postoperative CRP and neutrophilic granulocyte levels only a few days after initial surgical intervention, as well as slow decrease of absolute and relative CRP and neutrophile values are significant risk factors for failure of a single surgical intervention. The direct proportionality of absolute and relative changes in this study highlight the known physiological exponential course of CRP and neutrophil levels27,28. Consequently, it was sufficient to include one of the two progression parameters into the SYNC scores. The predictions were based on symptom duration, CCI, bacterial culture results, and absolute CRP and neutrophilic granulocyte changes by the third postoperative day, achieving an AUC of 0.80. Including additional information on the CRP change from admission to the fifth postoperative day further increased the AUC to an excellent 0.85. However, as there was not enough data available for internal validation, external validation is required before our scoring systems can be used in clinical practice.

In our cohort 65 patients (26%) had a suspected reinfection with consequent need for further surgical treatment. These numbers are comparable to recent studies, which described rates between 11 and 40%2,3,7,18,19,20,21,22,23. In terms of surgical technique, arthroscopy was the preferred method in case of knee infection (72%) and arthrotomy was chosen for the majority of shoulder infections (64%). No significant differences regarding failure rates were observed for knee infections. Though, we found a trend towards better results after arthroscopy, which is consistent with current literature, indicating that knee arthroscopy reduces reoperation rates while improving postoperative range of motion2. Shoulder arthroscopy was associated with a higher failure rate in our study, yet current evidence does not provide a clear indication of the superiority of one surgical method over the another. Failure rates in current literature range from 5.6 to 100%, and Abdelmalek et al. found higher reoperation rates in arthroscopy in comparison to arthrotomy for the treatment of native shoulder septic arthritis, Acosta-Olivo et al. report a tendency towards lower reinfection rates for arthroscopy, and Memon et al. did not demonstrate the superiority of either method3,7,12,18,29,30.

We observed a trend towards lower Gächter classes in patients treated with arthroscopy compared to arthrotomy, but no difference in Gächter stages regarding the risk of failure. This underlines previous findings, indicating that patients with a higher Gächter classification are more likely to benefit from arthrotomy than from arthroscopy, and appropriate choice of surgical procedure can help minimize the risk of failure31.

Our study yielded positive microbiological cultures of joint aspirate for almost half of all cases, twice as frequent and therefore significantly more common in the failure group, in which infection was confirmed histologically. The current body of evidence highlights that negative microbiological cultures occur in 12–50% of cases despite strong clinical suspicion of septic arthritis, especially if crystals or clotting are present20,21,22,32. The sensitivity of bacterial culture reportedly ranges between 75–95%, which might decrease in case of atypical organisms or antibiotic treatment before joint aspiration, and in 9–14% of cases the pathogen can only be detected through additional blood cultures33,34. This underlines the necessity for a more comprehensive approach to diagnosis including clinical, radiological, and histological evidence2,35.

In the presented cohort, Staphylococcus aureus was the most common pathogen and accounted for 49% of all organisms, similar to previous reports ranging between 37–56%2,19,36,37,38. Patients with Staphylococcus aureus infections showed no significantly higher failure rate in our study. However, current data on this topic is ambiguous: some studies found a significant association12,19,24, whilst others did not7,20. Different rates of MRSA infections might explain the contradicting findings. In our cohort, there were only four MRSA cases, and therefore too few to draw conclusions and far fewer than reported in comparable studies7,19,39. Geographical differences in the presence of MRSA might explain the inconsistent findings40.

The current study confirms previous observations that symptom duration and CCI are risk factors for failure of a single surgical intervention. However, none of the analyzed single comorbidities was associated with intervention failure3,7,22. Further, we found a significant difference of about 1 mg/dl in Hb levels at discharge or before reoperation between successful and failed single surgical intervention groups. This might be explained though a more radical synovectomy, and therefor increased blood loss, for patients suffering from a more severe infection, but a more detailed assessment of postoperative Hb levels is needed.

As of today, there exists no guideline on how to decide whether or when to repeat irrigation and debridement in patients with suspected persistent infection, and there is currently no uniform definition of treatment failure7.

The median time to second surgery was 15 days within our cohort, but for the majority of patients classified as high risk, repeated surgery was correctly predicted 10 days earlier than that. This indicates that levels and changes of infection parameters only a few days after initial surgery already need to be considered as risk factors for treatment failure, instead of drawing conclusions from them only after more extended periods of watchful waiting. Therefore, a faster decision for return to operating room, reinforced by score results, could potentially prevent further joint destruction, and preserve functionality. As for patients in the intermediate risk group, the indication for earlier and more frequent follow-ups may be made.

The current study has some limitations: First, although it is based on the largest dataset of risk factors for shoulder and knee arthritides in current literature, its sample size still is comparatively small. Including the two most commonly affected large joints on the one hand increased sample size, but on the other hand neglect joint specific pathophysiology or treatment response. Our data quality was also limited by the retrospective data collection, with consecutive incompleteness and missing randomisation. No functional parameters could be considered due to insufficient documentation, and information on the administration of antibiotics prior to hospital admission was not available. Second, as there were not enough data for internal validation, there is a clear risk of overfitting for our predictive scores, which needs to be taken into consideration when applying them in clinical practice. Third, in terms of surgical approaches, decisions were made based on the opinion of the consultant orthopedic surgeon.

In summary, the course of CRP and neutrophil granulocytes between admission and the fifth postoperative day, in combination with CCI, symptom duration, and bacterial culture results can help predict the need for further surgical intervention earlier and more accurately. Moreover, a decrease in hemoglobin and high Kellgren-Lawrence grading as factors significantly associated with more than one surgery.