Identifying risk groups of infectious spondylitis in patients with end-stage renal disease under hemodialysis: a propensity score-matched case-control study

Infectious spondylitis (IS), also known as pyogenic spondylitis, is a rare but life-threatening condition [1]. Clinical entities of IS include vertebral osteomyelitis, septic discitis, and epidural abscess. IS most commonly arises from hematogenous spread of bacteria, and often involves adjacent vertebrae and the corresponding intervertebral disks. IS is a disease affecting elderlies in their fifth decade of life with increasing incidence every decade thereafter, and occurs 1.5 times more frequently in male [2, 3]. Because most IS cases were believed to be predisposed by bacteremia, the reported risk factors were statuses related to suboptimal immunity, including old age, diabetes mellitus (DM), chronic kidney disease (CKD) and other immunocompromising conditions [4, 5]. In addition, any direct route into the bloodstream raises the risk of systemic bacterial infection. These risk factors synergistically contribute to the morbidity and mortality of patients with end-stage renal disease (ESRD). For instance, it has been reported that the annual mortality due to sepsis was approximately 100 to 300 fold higher in dialysis patients as compared to that in the general population [6]. Therefore, patients with ESRD under hemodialysis (HD) are under constant risk of developing IS and require clinical attention.

To diagnose IS in patients with ESRD under HD, clinical, radiological, and laboratory evaluations are often required. Early identification of IS to prompt initiation of therapy is vital in ensuring successful treatment and preventing morbidity. Because of the insidious onset of IS symptoms and the fact that most of the specific changes in imaging tests require two to eight weeks after the onset of the infection, the diagnosis may be delayed and lead to serious complications including irreversible neurological deficits and death [7]. Thanks to the high sensitivities of serum markers such as C-reactive protein (CRP) and erythrocyte sedimentation rate (ESR), they could serve as screening tests prior to magnetic resonance imaging (MRI), the gold standard to diagnose and evaluate the extent of IS [1]. However, whether certain serum markers provide a further predictive potential for a future IS event in these patients remains to be explored.

Our aim was to perform a 1:4 propensity score-matched case-control study among HD patients to identify the differences in baseline characteristics and potential serum markers between the patients who developed IS one month later and those who did not. Because the differences in age, gender, and DM status may affect blood tests results and clinical outcomes, we matched on these important confounding factors and adjusted them in our models (as presented below).

The study group contained 16 patients aged 20 years or older, who received regular HD and had been diagnosed with IS from 1 July 2006 to 30 June 2015. For each patient in the study group, 4 controls were matched by propensity score by age, gender, and status of DM. This matching was confirmed to be adequately balanced by calculating the level of SMDs of age, which was 0.0258.

The baseline characteristics according to the grouping are shown in Table 1 below. Patients between the study and control group had no statistical difference in age, gender, DM, hypertension, coronary artery diseases, heart failure, cerebral vascular diseases, liver cirrhosis, malignancy, and history of spine operation. Patients with IS had higher prevalence of immunosuppression status (6.3% vs. 0%, p = 0.04) and degenerative spinal disease (87.5% vs. 51.6%, p = 0.01). Patients in the study group had a shorter HD vintage (4.65 ± 4.36 vs. 13.44 ± 7.30, p <  0.001), a higher possibility of receiving HD through tunneled cuffed catheter (43.8% vs. 12.5%, p <  0.01) rather than arteriovenous fistula (37.5% vs. 68.8%, p = 0.02), and a higher likelihood of receiving access operation within 6 months (62.5% vs. 12.5%, p <  0.001). Patients in both groups had no significant difference in their primary cause of ESRD.

The blood investigation results of the study group one month before the diagnosis of IS showed significantly lower hemoglobin level (8.86 ± 1.3 vs. 10.2 ± 1.6, p <  0.001), higher RDW (16.3 ± 2.2 vs. 14.5 ± 1.4, p = 0.001), higher ALP (131.2 ± 84.5 vs. 83.4 ± 38.8, p = 0.001), lower albumin (3.35 ± 0.65 vs. 3.90 ± 0.38, p = 0.001), and higher hsCRP (33.1 ± 43.6 vs. 12.4 ± 23.9, p = 0.004) compared to the control group. On the other hand, leukocyte count, platelet, BUN, creatinine, potassium, calcium, phosphorous, ALT, glucose, and ferritin level were not significantly different between the two groups.

Regarding HD vintage, we found a significant association between it and a recent access operation (p = 0.032). In addition, we also identified positive relationships between HD vintage and both serum levels of albumin and hemoglobin (p = 0.003 and 0.042, respectively). To clarify its impact on our findings, we further performed the second propensity score matching model adding HD vintage as a matching variable and showed overall similar results as demonstrated in Table 2, with the exception being there were less underlying disease of hypertension (50% vs. 79.7%, p = 0.02) in the IS group.

Variables that were significantly different in patient’s baseline characteristics and clinical features were adjusted by sex, gender and DM in order to identify the risk factors for IS in HD patients (as shown in Table 3). In the first matching model, the following variables remained significant after adjustment: underlying disease of degenerative spinal disease (odds ratio [OR] = 12.87, 95% confident intervals [CI] = 1.89–87.41; p = 0.009), HD vintage (OR = 0.64, 95%CI = 0.51–0.81; p <  0.001), HD through a tunneled cuffed catheter (OR = 6.75, 95%CI = 1.74–26.14; p = 0.006), access operation within recent 6 months (OR = 13.27, 95%CI = 3.53–49.91; p <  0.001), hemoglobin (OR = 0.47, 95% CI = 0.29–0.76; p = 0.002), RDW (OR = 2.01, 95% CI = 1.33–3.04; p = 0.001), ALP (OR = 1.02, 95% CI = 1.00–1.03; p = 0.008), albumin (OR = 0.09, 95% CI = 0.02–0.35; p = 0.001), hsCRP (OR = 1.02, 95%CI = 1.00–1.04; p = 0.046). The inverse propensity score weighting used to estimate the effect of recent access operation on IS consistently demonstrated a significant association of recent access operation with an IS in the near future (OR = 11.52, 95% CI = 3.20–41.50; p <  0.001).

When it comes to the second matching model, generally similar results were shown in Table 4. The main divergences being hsCRP no longer remained significantly different, while underlying disease of hypertension was significantly less in the IS group after adjustment.

The ROC curves for the variables that remained significantly-related after adjustment, hemoglobin, RDW, ALP, and albumin, were shown in Figs. 2, 3, 4 and 5. For hemoglobin, the area under the curve was 74.5% while 9.35 was the cutoff level as a risk factor for IS (sensitivity: 62.5%; specificity: 75.0%). For RDW, the area under the curve was 77.5% while 15.65% was the cutoff level for RDW to point at a possible diagnosis of IS (sensitivity: 68.8%; specificity: 82.8%). When it comes to that of ALP, the area under the curve was 79.0% and 103.5 was the cutoff level as a risk factor for IS (sensitivity: 73.3%; specificity: 79.7%). In the case of albumin, the area under the curve was 77.8% while 3.73 g/dL was the cutoff level for albumin to indicate an increased risk for IS (sensitivity: 81.2%; specificity: 76.6%).

We managed to evaluate adjusted predictive models for future IS by binary logistic regression analyses with forward selection of the significantly-related variables after adjustment. The first matching model demonstrates that the two selected factors following adjustment were HD vintage and ALP, as shown in Table 5. No collinearity was found between the chosen variables, and this multivariate ROC curve had an area under the curve of 92.3%.

We also followed the same approach for the second matching model, turning out that the two selected factors following adjustment were RDW level and an access operation within recent 6 months, as shown in Table 6. No collinearity was found between the chosen variables, and the multivariate ROC curve had an area under the curve of 88.8%.

In addition, we analyzed lab data collected around three months before the diagnosis of IS and compared them to those of one month before diagnosis. The findings remained consistent and no apparent difference was noted (Additional file 1: Table S1).

In the first propensity score-matched model matched by age, gender and DM status, we found that patients with the history of a degenerative spinal disease, shorter HD vintage, recent access operation, or HD through a tunneled cuffed catheter was significantly related to the development of IS after adjustment. Moreover, levels of laboratory data, including lower hemoglobin, lower serum albumin, higher RDW, higher ALP, and higher hsCRP were also significantly linked to IS after adjustment. Since we found HD vintage to be different between the two groups, which may be confounding, we built the second propensity score-matched model with the addition of HD vintage as a matching factor. Overall, consistent results were noted even after adjustment. In addition, we further compared these laboratory findings documented about three months before the diagnosis of IS to those collected one month before diagnosis in light of the insidious nature of IS, and consistent results were still noted. In order to ensure timely diagnosis of IS in clinically-suspected HD patients, we also identified optimal cutoff-levels of serum hemoglobin, RDW, ALP, and albumin. To our knowledge, this is the first study to show relationships between these variables and the odds of IS among HD patients.

The significant relationships linking recent access operation and tunneled cuffed catheter to increased odds of IS plausibly suggest an increased risk of catheter access infection due to transient bacteremia in these patients on HD [8]. When it comes to laboratory studies, lower hemoglobin and albumin levels have long been linked to inflammation and malnutrition. It is well-known that HD poses an increased risk of inflammatory syndromes in patients through both patient-related and HD-related factors [9]. This chronic inflammatory effect had been consistently reported to be related to decreased serum albumin level [10] and subsequently increased mortality [11]. In addition, malnutrition also increases the mortality in patients on HD [12, 13], likely through decreasing the number of T cells and impairing the production of cytokines [14]. On the other hand, the associations of both increased RDW and increased ALP with the mortality of HD patients have both been reported [15‐17]. Elevated RDW may reflect multiple underlying conditions, including ineffective erythropoiesis, malnutrition, increased oxidative stress, and endothelial dysfunction [18]. Although the increase in ALP has been reported as a reliable marker for the severity of the high-turnover osteodystrophy in patients on HD [19], and is associated with renal osteodystrophy and vitamin D deficiency [20], this is an independent risk factor for altered immunity and inflammation [21]. As for hsCRP, both of its indicative role of inflammatory status and its prognostic part in HD patients have been well-recognized [22‐24]. Our result demonstrated that elevated levels of these laboratory findings should raise clinical concerns for the inflammatory process linking to an increased risk for IS in addition to the HD status.

After clarifying the adjusted odds ratios for each variables for IS, we further managed to generate a multivariable model with adjustment to look into the combinations of factors that predicted IS events. Within model 1, we noted that shorter HD vintages and elevated ALP were selected as the most significant factors. While within model 2, we found recent access operation and serum RDW level as the two most significant factors without collinearity in a single regression model. Both combinations have adequate predictability for IS within future 1 month. Due to the natural rarity of IS [1, 7], there was limited sample size and it may be improper for us to add in more variables in light of it. However, based on our findings, we strongly suggest physicians to maintain a high degree of suspicion for infectious process facing symptoms of back pain, neurological deficit, unexplainable fever, weight loss, lethargy, or confusion in patients receiving maintenance HD, particularly if there has been an unexpected high ALP level, or inexplicable low RDW level following a recent access intervention in patients with relatively short HD vintage.

Intriguingly, we found that the study group had a shorter mean HD vintage before we included it as a matching variable. This could be partly explained by the fact that patients require a recent access operation to start hemodialysis, and a recent access operation is a significant factor for increased odds of IS. This significant relationship between shorter HD vintages and a recent access operation was indeed observed in our study, while adding HD vintage as an additional matching variable did not affect the significant relationship between a recent access operation and IS. In other words, factors besides recent access operation that relates to shorter HD vintage, such as the inexperience in self-care, may also contribute to this finding. Moreover, though longer HD vintages were reported to be related to declined nutritional status and increased mortality [25], our result showed a positive relationship between HD vintage and some markers of nutritional status, including serum albumin and hemoglobin levels (data not shown). These findings could be explained by the fact that all patients included were regularly followed up and treated in a tertiary medical center, and their nutritional status was closed monitored throughout the hemodialysis course. As a result, the better nutritional status over time may also explain the decreased odds of IS as the HD vintage gets longer. We believe two of the reasons discussed above both contributed to our findings because the HD vintages in the two study groups were still relatively shorter after stratification by recent access operation.

The limitations of our study include its observational design and the small effective sample size. Future studies with a greater sample size or a large prospective cohort design are warranted to confirm our findings.

In conclusion, we found that HD through catheter, history of degenerative spinal disease, immunosuppression status, or recent access operation, low hemoglobin or albumin levels, high RDW or ALP levels were associated with the subsequent diagnosis of IS in patients undergoing HD, facilitating early detection of the risk group. These findings also highlight the potential pathogenesis of blood access and malnutrition as a link between HD patients and IS, and whether improving these correctable factors could prevent HD patients from IS remains a subject for future studies.

Based on this PSM study, we found that patients with shorter HD vintage, a recent access operation history, or HD through a tunneled cuffed catheter was significantly related to the development of IS after adjustment. In addition, lower hemoglobin, lower serum albumin, higher RDW, and higher ALP levels, were significantly linked to IS after adjustment. Concurrent presentations of the often unspecific IS symptoms with these risk factors in patients under maintenance HD should prompt clinical awareness towards IS, especially when unexpected high ALP levels, or inexplicable low RDW levels following a recent access intervention were found in patients with relatively short HD vintages.