Pneumatosis Intestinalis and Porto-Mesenteric Venous Gas: A Multicentre Study


 Background: Estimating the prognosis of patients with pneumatosis intestinalis (PI) and porto-mesenteric venous gas (PMVG) can be challenging. The purpose of this study was to refine prognostication to improve decision making in daily clinical routine. Results: A total 290 patients with confirmed PI were included in the final analysis. The presence of PMVG and mortality (90d follow-up) were evaluated with regard to the influence of possible risk factors. Furthermore, a linear estimation model was devised combining significant parameters to calculate accuracies for predicting death in patients undergoing surgery by means of a defined operation point (ROC-analysis). Overall, 90d mortality was 55.2% (160/290). In patients with PI only, mortality was 46.5% (78/168) and increased significantly to 67.2% (82/122) in combination with PMVG (median survival: PI: 58d vs. PI and PMVG: 41d; p<0.001). In the entire patient group, 53.5% (155/290) were treated surgically. OS was >90d in 25% (72/90) of conservatively treated patients. In these cases PMVG/PI and was defined as “benign”/reversible. PMVG, COPD, sepsis, and low platelet count correlated with a worse prognosis. To identify patients deteriorating despite surgery, our calculation model reaches accuracies of 97% specificity, 20% sensitivity, 90% PPV and 45% NPV. Conclusion: Although PI is associated with high morbidity and mortality, „benign causes” are common. However, in concomitant PMVG, mortality rates increase significantly. Our mathematical model could serve as a decision support tool to identify patients who are least likely to benefit from surgery, and to potentially reduce overtreatment in this subset of patients.


Introduction
Pneumatosis intestinalis (PI) is de ned as the presence of gas within the submucosal or mucosal wall of the intestine (1). PI was rst described anatomically in 1730 by Duvernoy and named nearly hundred years later by Mayer as pneumatosis cystoides intestinalis (1)(2)(3). PI is a radiographic or histopathologically veri ed phenomenon, rather than a clinical diagnosis with potentially life-threatening consequences for the patient. The aetiology of PI is complex. But overall, three causes have been identi ed as the source/origin of gas within the intestinal wall: 1: gas formed in the intestine, i.e. luminal gas (4).
Aetiology may range from ischemia or bowel obstruction to sepsis and many other conditions. Apart from these clearly life-threatening pathologies, benign / reversible causes like respiratory disorders and / or unfavourable drug related side effects -e.g. chemotherapy-associated -have been reported (10)(11)(12). Clinical presentation can be unspeci c, but the most common symptoms involve abdominal pain, diarrhoea, bloody stools, fever, and shock (1,13,14). Upon presentation, a clinical examination must be performed, and abdominal x-ray can be helpful. However, in clinical reality, computed tomography (CT) is the modality of choice for diagnosing PI (3,15).
Yet, the role and prognostic relevance of gas ascending into the liver, causing porto-mesenteric venous gas (PMVG) remains unclear. PMVG has been both described as linked to increased mortality as well as a reversible phenomenon (1,10,(16)(17)(18)(19) (Fig. 1). For most "unexperienced" radiologists and surgeons, the pattern of PI and gas in the portal venous system is indicative of a potentially irreversible intestinal infarction and associated with a poor prognosis. Although PI and PMVG have been discussed for decades, there is little to no comprehensive evidence-based data on PI and especially PMVG. Most data available on benign and reversible causes of PVMG is based on case reports or case series (10,12).
Patients with PI and PMVG can have distinctly different prognoses. In clinical practice, it can be challenging to decide upon further treatment of patients with PI, even for experienced clinicians. The purpose of this study was to re ne prognostication for these patients to improve decision-making in daily clinical routine with a special focus on patient outcome after surgical treatment.

Methods And Material
Patients: Our institutional review board approved this retrospective multicentre study (registration number EA2/238/17). The study protocol conforms to the ethical guidelines of the 2002 Declaration of Helsinki.
Informed consent was waived due to the retrospective nature of this study. The study was conducted at three tertiary care university medical centres (University Medicine Berlin, University Medical Centre Mainz, and University Hospital of Cologne). The respective radiological information systems (RIS) were searched for reports led during the timespan of January 2008 to January 2018, and contained the following keywords: pneumatosis, pneumatosis intestinalis, gas in the bowel wall, venous gas, gas in the portal vein and intramural gas (Fig. 2). Where available, pathology and surgery reports were retrieved from the hospital information systems. The diagnosis of PI was con rmed by a board-certi ed radiologist after rereading the respective imaging studies. To increase quality and to avoid false positives, all cases were reread by two radiologists in consensus at every centre; one of each with at least ten years of abdominal CT experience (University Medicine Berlin: D.G and M.D.S; University Medical Centre Mainz: R.K and F.H; University Hospital of Cologne: D.P.D.S and J.H). Re-reading was performed to con rm the presence of gas within the gastrointestinal wall, as opposed to luminal gas, i.e. pseudo-pneumatosis phenomenon (false positive).
Mortality and overall survival (OS) were evaluated. OS was determined within a 90-day follow-up after the initial CT examination. To control for selection-bias, the following qualitative and quantitative risk factors were evaluated:

Qualitative parameters
Age, sex (male and female), presence of PMVG, sepsis, vascular disease de ned by varying degrees of stenosis (50-75%, 76-100%) of the superior or inferior mesenteric artery (AMS and AMI) or the coeliac trunk, underlying oncologic diagnosis, COPD, cardiovascular disease, presence of bowel obstruction (large vs. small bowel obstruction), decreased bowel wall enhancement and the therapeutic use of steroids.

Quantitative parameters
Length of the PI-affected gastrointestinal segment (from 0-100cm in increments of 10 cm), laboratory parameters including CRP, procalcitonin (PCT), platelet count, and lactate.

Imaging acquisition
All patients were examined with multislice CT scanners and underwent single or multiphase CT examinations (arterial, and/or portal-venous and/or venous/delayed phase). At the rst institution (University Medicine Berlin) patients were either examined using a 128-or one of two 64-multislice CT scanners (Revolution HD and Revolution EVO, GE Healthcare, Milwaukee, WI, USA) (Somatom Sensation 64, Siemens Medical Systems, Forchheim, Germany). All patients received a non-ionic intravenous contrast agent (either Ultravist 370, Bayer Schering, Berlin, Germany, Xenetix 350, Guerbet, Villepinte, France or Imeron 400, Bracco, Milan, Italy). At the second institution, patients were examined with either a 256-multisclice CT scanner or by a 64-multislice scanner (Philips iCT and Brilliance, Phillips, Amsterdam, Netherlands). All patients received a non-ionic intravenous contrast agent (Imeron 400, Bracco, Milan, Italy). At the third institution, patients were examined with either a 256-multisclice CT scanner (Philips iCT, Philips, Amsterdam, Netherlands) or a 128-multislice scanner (Phillips IQon, Philips, Amsterdam,
Of the patients with a mechanical bowel obstruction, 42% (18/43) had small bowel and 58% (25/43) large bowel obstruction. Mechanical bowel obstruction was observed signi cantly more often in the "PI only"-subgroup (p = 0.004), while the location (small vs. large bowel) did not differ signi cantly. Sepsis was documented in 26.9% (78/290) of patients. Surgery was performed in 53.5% (155/290) of all patients (Fig. 1). Sepsis was signi cantly more prevalent in patients with coinciding PMVG (p = 0.010), compared to the "PI only"-group, all other distributions showed no signi cant differences (Table 1). Netherlands). All patients received a non-ionic intravenous contrast agent (Accupaque 350, GE Healthcare, Milwaukee, WI, USA). Scan direction was craniocaudal in all institutions.

Statistical analysis:
All statistical analyses were performed using XLSTAT (Version 2011.0.01; Addinsoft SARL, New York, USA) and / or with R 3.5.1 using RStudio 1.1.463; R Core Team (2018) (20). Survival plots / Kaplan-Meier Curves were calculated and designed using R 3.5.1 with RStudio and help of the ‚survival'-package (21). Contingency tables were used for statistical results with proportional distributions. OS was estimated using the Kaplan-Meier analysis and compared using the Logrank test. The Cox proportional hazard model was used for multivariate analyses to evaluate the in uence of each risk factor on overall survival.
A multivariate logistic regression model was used to test for in uence of each risk factor on dichotomized variables (overall mortality, "benign" PI etc.). A p-value of less than 0.05 was considered statistically signi cant. Finally, a generalized linear / logistic regression model was built using the subset of patients who did not undergo surgery. We devised a statistical calculation model to predict death in surgically treated patients. A random sample of 100 patients from the conservatively treated cohort was used to t the model. The remaining 31 conservatively treated patients were used to verify the model's performance. We then applied the model to the surgically treated patient cohort. We set an operating point for prediction of death with greater than 95% speci city to identify patients with an increased mortality risk despite surgical treatment. Accuracies were given as sensitivity, speci city, positive and negative predictive values (PPV/NPV).  (Fig. 3). and survived the 90d follow-up (n = 72) ("benign" cohort). The second group included patients who were un t for surgery and died within the 90d follow-up (n = 63) ("deceased" cohort). Median survival for patients who died within the 90d follow-up was 9d ± 2.1 (SD; 95% CI 4.7-13.1). The prevalence of PMVG was signi cantly higher (p < 0.05) in the "deceased" cohort 52.2% (33/63) compared to the "benign" cohort (27%; 19/72) (Figs. 2 and 3). Detailed results can be found in Table 1.

Comprehensive risk pro le analysis
In a second step, the possible in uence of several risk factors on survival were calculated by means of multivariate regression models ( Table 2).  Table 2).

Multivariate logistic regression model
The following parameters were associated with reversible / "benign" PI: the absence of PMVG (p < 0.042), the absence of COPD (p < 0.017) and low CRP levels (p = 0.006). All other risk factors had no signi cant in uence on reversibility of PI in the conservatively managed group. A low platelet count was found to be associated with fatal outcome with the group of patients who were managed conservatively/i.e. nonsurgically (p < 0.001).

Multivariate estimation model
As abovementioned, the following parameters were statistically signi cant for a decreased OS: PMVG, COPD, sepsis and low platelet count.
After subdividing patients into two cohorts, a surgically treated and conservatively treated cohort, in the no-surgery cohort, the parameters were re-tested again by means of a training set to predict death. Pvalues generated from the training set were as follows: PMVG: p = 0.023, COPD: p = 0.083, sepsis: p = 0.006 and thrombocytes: p = 0.003 -AUC: 0.818 (Fig. 4a). In a second step, we used the model to analyse the surgery-cohort with these parameters to predict death and calculated an AUC of 0.689 (Fig. 4b). We identi ed and de ned a point on the curve with a speci city of 95%. To reach this speci city a sensitivity of 20% had to be accepted with a threshold of 1.8 (Fig. 4c). Accuracies generated by this model to predict death were as follows: 97% speci city, 20% sensitivity, PPV 90%, NPV 45%. When dividing the cohorts according to the calculated threshold, resulting Kaplan-Meier curves also showed signi cant differences in overall survival (p < 0.001) (Fig. 4d).

Discussion
Prognostication in patients with PI and PMVG is challenging. This study con rms PI as a life-threatening diagnosis with an associated mortality rate of 54.1%. In the presence of PI and concomitant PVMG, mortality increases to 68.2%. Especially the presence of infection and ischemia correlate with a worse prognosis. Nevertheless, patients with "benign", i.e. reversible PI are common (25% in this study). The mathematical model we introduce helped us to pinpoint risk factors (PMVG, COPD, platelet count, sepsis) to identify patients with an increased mortality risk.
While the true incidence of PI is unknown, PI is diagnosed in day-to-day radiologic routine and its prognostic value remains unclear; especially in a setting with coincident PVMG. Some case reports have been published claiming that PMVG might have reversible causes, while others suggest PMVG has a dismal prognosis (22). The physical examination of the patients plays a key role in management and triage, as patients with signs of peritonitis or suspicion of intestinal ischemia and perforation are candidates for explorative surgery. However, other parameters to assist clinical decision-making and prognostic factors to potentially differentiate fatal from benign/ reversible PI are currently lacking. There are some factors, such as elevated serum lactate that seem to be associated with an increased mortality, mainly due to the worse outcome in patients with ischemia, albeit with a low level of evidence (Figs. 5 and 6) (23-25).
With 290 patients, this multicentre trial is one of the largest of its kind since the introduction of CT scanners.  (22). In our study, the pathology report was used as the gold standard for ischemia, which we then used to correlate ischemia with the presence of PMVG.
Despite the high mortality rates reported in this study, and the fact that survival was even lower in the presence of PMVG, we identi ed 80 patients who survived with non-surgical management. Although the prevalence of PMVG in this subgroup was low (10%), it was diagnosed in a few cases. This suggests that, PI in coincidence with PMVG can also have benign and reversible causes, and both radiologists and surgeons should be familiar with these (15), highlighting the need for further large-scale studies to identify the aetiology of "benign" / i.e. reversible PI and PMVG. Subsequently, we evaluated a comprehensive risk pro le for each patient retrospectively to understand and investigate factors associated with increased mortality as opposed to "benign"/reversible cases. First, we analysed factors associated with decreased survival rates. In line with the literature, we found elevated serum lactate levels to be associated with a worse prognosis (23)(24)(25). Old age, sepsis and an arterial stenosis were also found to correlate with shortened OS. Hence, we conclude that intestinal infarction, which is also linked to ischemia, remains the most life-threatening diagnosis in case of radiologically con rmed PI. Surprisingly, the strongest statistical correlation between shortened survival and the only signi cant factor to occur in patients who deceased without surgery, was a low platelet count. Platelets play a central role in the regulation of blood ow and thrombogenic cascades (27). An increased platelet count was thought to be directly associated with increased vascular in ammation and thrombogenicity and therefore to correlate with thrombotic events like ischemic stroke and myocardial infarction (27,28). Conversely, in this study, a low platelet count was associated with a worse prognosis. It is possibly even more interesting to investigate factors that correlate with "benign" PI to identify those patients who do not require surgery but conservative treatment. In our study, low CRP levels, the absence of PMVG, sepsis and COPD were predictive for reversible PI.
When combining the signi cant parameters identi ed in the logistic regression model (PMVG, COPD, sepsis and low platelets) to predict a fatal outcome in surgically treated patients, our multivariate model reached a speci city of over 90%. This statistically validated risk pro le could serve as a decision support tool to identify patients who are unlikely to bene t from surgery, and thus potentially reducing overtreatment.

Limitations:
Our study has several limitations. First and foremost, the retrospective study design and the fact that all readers were aware of this design, introduce a potential detection bias. Secondly, there is an inherent selection bias, as the RIS were searched for reports containing PI and PI-related terms. Hence, if none of these terms were mentioned in the radiology report, the patient was not included. As a diagnosis of PI does not require an arterial contrast phase, not all of the included patients received a CT scan comprising that phase. This can limit the chances of detecting decreased bowel wall enhancement and therefore explain why ischemia did not signi cantly correlate to the pathology results in all patients undergoing surgery. Also, the medical history was incomplete in some patients. Although pathological reports were available in surgically treated patients they did not provide 100% speci city, which is a further potential cause of distortion.

Conclusion
PI is not necessarily associated with intestinal ischemia but may be attributed to various other causes. The coincidence of PI and PMVG remains indicative of intestinal ischemia and strongly correlates with signi cantly increased mortality rates. The validated risk pro le we introduce can serve as an ancillary decision-making tool to identify patients who are unlikely to bene t from surgery and to potentially reduce overtreatment. • Consent for publication: All authors red the manuscript and agreed for publication.
• Availability of data and materials: The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.