Prognostic value of miliary versus non-miliary sub-staging in advanced ovarian cancer

Abstract

Objective

The presence of miliary disease during initial ovarian cancer debulking may reflect a distinct mode of peritoneal spread independent from size-based tumor staging and may explain variation in response to treatment and survival outcomes. To infer the prevalence, presentation and clinical implications of miliary disease we reviewed existing surgical records.

Methods

Reports were available for 1008 primary debulking surgeries for ovarian, primary peritoneal or fallopian tube cancer between 2001 and 2015 (685 reports from 2005 to 2015). Clinical outcome data was available for 938 patients. We analyzed a high-stage sub-cohort for survival (N=436).

Results

Most records were evaluable for miliary disease (761/938); for these, the miliary phenotype was highly prevalent (249/761, 32.7%) and often accompanied by ascites (185/249, 74%). While optimal debulking rates were unaffected by miliary disease, total resection (R0) rates were poorer. Liver, stomach, spleen or bladder appeared to be sporadically involved while the omentum, mesentery, bowel, peritoneum and diaphragm were affected simultaneously (Spearman rho > 0.5). Overall, miliary disease was associated with worse progression free survival, overall survival, and time from relapse to death independent of stage. Survival effects were particularly strong for Stage IV disease where median overall survival varied by over 30 months (log-rank p=0.002).

Conclusions

Miliary disease is an identifiable surgical phenotype reflecting a distinct clinical trajectory that adds prognostic information to standard disease burden-based staging. These findings should permit further retrospective investigation in a wider cohort and prompt the consideration of prospective structured operative reporting standards and treatment strategies.

INTRODUCTION

Surgical staging represents one of the strongest prognostic biomarkers for advanced ovarian cancer (1). While there are multiple staging schemas (2, 3), all systems share a site and size based classification where greater disease burden and spread into the peritoneal cavity define higher stages.

Recognizing the heterogeneity in the involvement of specific abdominal organs, Fagotti and colleagues developed a laparoscopic model for predicting achievement of optimal cytoreduction including mesenteric, bowel, stomach, and liver involvement as highly-specific, primary risk factors (4). Surprisingly, their recent study noted that peritoneal/diaphragmatic carcinomatosis and omental caking were more accurate predictors for sub-optimal cytoreduction than degree of spread and tumor size (5).

These factors may reflect a miliary-type disease defined by the presence of diffuse spread of sub-centimeter nodules or plaques across multiple organ/peritoneal surfaces resembling tuberculosis peritonitis (6). This pattern of disease spread is distinct from non-miliary disease in which the growth pattern is large bulky disease with few peritoneal implants demonstrating an exophytic growth pattern (Figure 1).

Figure 1.

Representative laparoscopic views showing tumor bulk above retracted mesentery (A), pooled ascites below peritoneal miliary disease (B) and a representative gradient of miliary disease to confluent plaque again with ascites (C).

Recognizing that despite many surgical and critical care advances, significant variations in debulking rates remain (7, 8), it is tempting to speculate that miliary disease is both the source of suboptimal cytoreduction (4) and a major predictor of poor clinical outcomes (9). Deliberate intervention to improve total resection (R0) rates (10) suggests that moving debulking goals from resecting all bulky disease (optimal) to achieving R0 may or may not improve survival (11, 12) and the difference in residual disease size roughly fits the miliary component.

It may be an oversimplification to consider simply the size and location of intra-abdominal metastases. Auer and colleagues have presented evidence in a medium number of cases that miliary disease is a molecularly distinct subtype of metastasis (11, 13). Because cases can present with miliary-type spread alone or in combination with heavy tumor burden, and the latter are upstaged algorithmically, there might be important to sub-stratify heavy burden tumors with and without miliary disease.

To understand the clinical trajectory associated with miliary versus non-miliary disease, we undertook a retrospective study of surgical records from patients undergoing maximal debulking surgery for ovarian cancer. We reviewed these patients’ surgical and adjuvant chemotherapy outcomes as well as time to progression or relapse and time following relapse to death. Our objective was to compile outcomes data to evaluate whether a miliary disease constitutes a distinct clinical phenotype.

MATERIALS AND METHODS

Patient Cohort

Following institutional review board (IRB) approval, we identified 1008 records of primary debulking surgeries with matching surgical pathology records for epithelial ovarian cancer conducted by the gynecologic oncology service at Roswell Park Cancer Institute between January 2001 and December 2015. Of these, n=938 patients were eligible for evaluation (20 were recorded as not cancer, 70 were determined to have a non-qualifying primary site after surgery). No patients received neoadjuvant chemotherapy. Of these, 177 operative reports were excluded because they lacked sufficient detail to determine the extent of disease; we were able to assess the pattern of disease spread in 81% of cases (761/938).

Surgical record review

We developed a set of variables and a protocol for standardized abstraction: miliary disease was defined by the direct mention of miliary disease, the description of small nodules and/or plaques consistent with the images used in the MITO-13 study (14). We trained three physician abstracters (DC, AK, LA) using these images and they scored individual surgical records for the presence of miliary disease, organ site involvement, and ascites. Independently, we applied a natural language processing algorithm that identified context relevant mention of nodular disease. The automatic and human scoring agreed in 78% of the cases with 62% specificity and 85% sensitivity (using human scoring as the gold standard). We estimated moderate inter-rater reliability for a random subset of 666 overlapping cases (Kappa=0.46, p<0.001). The rate of miliary disease scoring was similar across raters (p=0.726) and the rate of scorable records was similar (p=0.449). Miliary disease was consistently defined across year of diagnosis with no change in relative frequency (chi-square test p=0.30). Using the set of Stage I through IIIA cases as a negative control, we note that only 25 were incorrectly scored (2.4% overall) and a handful were subsequently upstaged.

Clinical Correlates

We adhered to the FIGO 2014 staging scheme throughout the study (2). We summarized patient characteristics in Table S2, which details the standard clinicopathologic and prognostic variables by stage. Optimal debulking is defined as resection to below 1cm residual disease.

Response to initial treatment was considered complete if, after completion of adjuvant chemotherapy, there was both complete chemical (as measured by CA 125 < 35 U/mL) and radiographic (as measured by CT scan) resolution of disease. Response to treatment was partial if either chemical or radiographic parameters did not normalize after the initial adjuvant therapy. The clinical response was considered progressive disease if either chemical or radiographic parameters were increased from the baseline value after initial adjuvant therapy.

Progression-free survival was measured from the date of definitive cytoreductive surgery until first evidence of disease recurrence, defined by i) radiographic according to RECIST criteria (15), ii) chemical, with CA 125 rising to twice the upper limit of normal, or iii: tissue biopsy positive for disease. We defined progression for patients with partial response or progressive disease by i: radiographic evidence of progression according to RECIST criteria or ii: chemical progression with CA 125 two-times the nadir value. We defined overall survival as the date of definitive cytoreductive surgery until death. We censored patients not experiencing recurrence, progression, or death at the date of last clinical contact.

Statistical Methods

All analyses were performed in R3.3.1 using the survival package. Tests were two-sided and specific tests are described within the results. Survival analyses use the Kaplan-Meier survival function to estimate medians and the log-rank test to compare the survival curves. Summaries of the regression models by organ involvement employ the restricted mean survival curve (16) fit to patients from each stage separately with the corresponding 95% confidence interval.

RESULTS

Correlates of miliary disease at primary debulking surgery

Across stages, miliary disease was moderately prevalent (226/556, 40.6%). The miliary disease rates were 42.1% (8/19) in stage IIIB, 42.5% (156/367) in stage IIIC and 55.1% (38/69) in stage IIIC/IV (163 patients lacked staging information). Miliary disease often presented with ascites (Table S2, 47.6%) but ascites was present in many patients without miliary disease (52%, 204/293); In 121 patients we were unable to assess the presence of ascites. Ascites greater than 2L appeared in all stages (Table S3) and miliary disease was associated with an average of 710mL additional ascites volume in high stage disease (stage adjusted linear model, p=0.0014).

We subsequently consider stage IIIC and IV cases with miliary or non-miliary disease (N=436). While there was no association between miliary disease and primary site (Table 1, chi-square test p=0.06), within stage, miliary disease was not associated with histological grade (p=0.05) or non-serous histologies (p=0.40).

Table 1.

Correlates of miliary disease at primary surgery (Low stage omitted)

	Stage IIIC Not Miliary	Stage IIIC Miliary	Stage IV Not Miliary	Stage IV Miliary	p-value
N	211	156	31	38
Age, mean	63.1	63.2	62.0	65.3	0.71
Primary Site					0.06
Ovary (N=703)	52%	33%	8%	7%
Primary Peritoneal	40%	43%	5%	13%
(N=149)	50%	21%	21%	7%
Fallopian Tube (N=40)	33%	53%	0%	13%
Multiple (N=46)
Grade: poorly differentiated or undifferentiated	75%	85%	87%	73%	0.05
Serous Histology	77%	83%	84%	74%	0.40
Ascites
Present	53%	84%	32%	84%	< 0.001
Median Volume (mL)	500	2000	200	1500	0.02
Volume > 2000 mL	32%	44%	13%	39%	0.05

Low stage cases are omitted.

Surgical debulking and miliary disease

While miliary disease was not associated with a lower rate of optimal cytoreduction (chi-square test p=0.763), stage IIIC disease was more often optimally cytoreduced than stage IV (75% versus 58%, t-test p=0.008). Conversely, while stage was not associated with R0 resection (p=0.39), miliary disease strongly correlated with sub-R0 resection (miliary 92.2% sub- R0 versus non-miliary 79.2%). This suggests that miliary disease may be directly contributing to poor surgical outcomes.

To understand whether site involvement varied between miliary and non-miliary cases, we tabulated reports of disease on organ sites throughout the abdomen (Table S4). Cases involving the diaphragm (44%) and were least likely to reach R0 cytoreduction (OR=0.73, 95% CI: 0.69–0.77) and most likely to show miliary disease (OR=1.76, 95%CI: 1.65–1.87).

Adjuvant Chemotherapy response and miliary disease

Following surgery, we observed a distinct clinical trajectory for high-stage patients with miliary disease versus non-miliary disease in Stage IIIC and IV patients (Table 2). At the completion of adjuvant chemotherapy, non-miliary patients had significantly higher rates of complete response (68.8% non-miliary versus 51.9%, t-test p=0.008) suggesting that miliary cases are unusually likely to persist or progress while on treatment and that suboptimal miliary cases may not be adequately cleared by systemic therapy. We observed that platinum sensitivity was similarly reduced by the presence of miliary disease 44.4% (Stage IIIC) and 23.5% (Stage IV) compared to 65.7% and 55.6% in non-miliary.

Table 2.

Clinical outcomes by miliary disease and staging

	Stage IIIC Not Miliary	Stage IIIC Miliary	Stage IV Not Miliary	Stage IV Miliary	p-value
N	211	156	31	38
Debulking Surgery
% Optimal	73.5%	77.6%	61.3%	55.3%	0.02
% R0	22.4%	6.5%	9.7%	13.2%	<0.001
Chemotherapy Response*	68.8%	51.9%	63.6%	43.5%	0.02
Platinum Sensitive	65.7%	44.4%	55.6%	23.5%	0.002
Months Median Survival
Progression Free	22.8	18.7	30.9	12.8	<0.001
Survival after surgery	48.7	30.7	50.6	15.2	<0.001
Survival after relapse	33.1	25.6	34.7	14.0	0.152

^*RECiST Evaluation complete response after end of adjuvant chemotherapy Low stage cases are omitted.

Progression, survival and miliary disease

Accounting for stage, miliary disease was uniformly associated with poorer prognosis (Figure 2): overall survival was 18 months shorter (Stage IIIC, log-rank p<0.001) and 35 months (Stage IV, p=0.002) shorter. The time from surgery to disease progression was also shortened (Stage IIIC, 4 months, p<0.001; Stage IV, 18 months, p<0.001). Importantly, survival after relapse was shorter for miliary disease in stage IIIC cases (by 8 months, p=0.028, N=152 cases with relapsed disease). Stage IV cases matched the shortening trend (20 months, N=17) but lacked the proper sample size to evaluate significance.

Figure 2.

Prognosis following surgery for patients with miliary (orange) versus non-miliary (blue) ovarian cancer stratified by stage shows miliary disease is a strong predictor of poor clinical outcome.

Adjusting for stage, age, histology, complete resection, and year of diagnosis, miliary disease increased the hazard of death (adjusted HR=1.90, 95%CI: 1.53–2.37, score test p<0.001), progression-free survival (1.62, 1.32–1.99, p<0.001) and death after progression (1.56, 1.11–2.21, p=0.011).

Multivariate site involvement and prognosis

To characterize multi-site involvement, we considered the correlation between reported organ sites (Figure 3A) and noted that the distinct organs -- liver, stomach, spleen, and bladder -- were weakly correlated implying they are involved independently of one another. In contrast, the peritoneum, omentum, mesentery, bowel, and diaphragm were often simultaneously involved (pairwise Spearman correlation > 0.5). The degree of independent organ involvement (liver, stomach, spleen, and bladder) was strongly associated with miliary disease (Figure 3B).

Figure 3.

Correlations between organ site involvement and prevalence (A) with miliary disease (B) affect overall survival (C), progression-free survival (D) and time after relapse (E). Organ involvement counts only the liver, stomach, spleen, and bladder.

Overall survival was incrementally worse for every involved organ site (Figure 3C, HR=1.16, 95%CI: 1.02–1.32) as was progression free survival (Figure 3D, HR=1.19, 95% CI: 1.05–1.34). The time from recurrence/progression to death was much more sensitive to organ involvement in IIIC cases (Figure 3E, HR=1.26, 95%CI: 1.03–1.53).

DISCUSSION

This chart review study suggests that miliary disease represents a distinct surgical ovarian cancer feature whose importance may have been overlooked in the presence of large bulky disease. About a third of cases presented with miliary disease and, independent of stage, miliary disease was a strong prognostic factor reducing survival by half. Observing fewer R0 resections in those with miliary disease is consistent with clinical experience and suggests that miliary disease may be the source of worse surgical outcomes and subsequent poor response to chemotherapy. This may explain the more rapid progression and shorter overall survival for these patients. Clinically, it is appealing to consider this surgical sub-type because it can be identified during a diagnostic laparoscopy or prior to adjuvant chemotherapy initiation.

A great deal of attention has been paid to disease burden at the time of clinical presentation with the goal of predicting optimal primary debulking surgery (PDS). Methods of pre-operative evaluation have been described including CA 125 levels (17), radiographic imaging (18), and laparoscopic evaluation of disease extent (4, 10). Two prior studies of neoadjuvant chemotherapy (NACT) have identified improved R0 resection rates with the use of neoadjuvant therapy, but without a difference in PFS or OS (19, 20). However, neither of these studies required pre-operative laparoscopic evaluation. In the SCORPION trial, a phase III randomized clinical trial comparing PDS and NACT in advanced epithelial ovarian cancer, patients underwent staging laparoscopy before randomization (21). Although survival data are not yet mature, we expect the results of the SCORPION trial will be informative regarding the impact of chemotherapy on miliary disease.

On the other hand, at least one large secondary analysis (12) has shown that PDS to R0 does not directly affect PFS or OS outcomes. Our finding that the locations of disease spread was correlated but not aliased with miliary presentation may elucidate the confounding factor. Our data suggest that miliary disease sets patients down a unique clinical trajectory. Patients with miliary disease are less likely to respond to platinum-based chemotherapies and once relapse is diagnosed, miliary disease portends poor response to salvage therapies.

It was evident from our analysis that cases of miliary or bulky disease were not mutually exclusive; it is highly likely that a case can have co-incident bulky and miliary disease. This is consistent with the distribution of predictive index scores presented by Fagotti and colleagues (4, 21). We have used staging as a crude surrogate for bulky disease and have observed that miliary disease may compound the burden of disease. For future research, we propose that surgeons record the presentation of miliary disease prior to resection.

Our conclusions are limited by the retrospective study design. In particular, because no standardized scoring or reporting was available, we have relied on human readers to abstract disease pattern. Despite this limitation, we report that it was possible to infer miliary patterns in the majority of case reports. While strengths of the study are the size and duration of our review, surgical goals and treatments have changed over time. We noted that year of diagnosis significantly affected prognosis but did not alter the effect of miliary disease. While seven attending oncologists dictated the operative reports throughout the study period and may have had different definitions of miliary disease, analyzed together, these potentially different definitions still permitted strong inference. We have also shown that a medical record review study is feasible and other investigators might review their own surgical records to independently verify our findings. We have not addressed the role of neo-adjuvant chemotherapy (excluded in our study cohort), but look forward to the SCORPION trial (22) results to address that setting. In this study, we identified potential strata within advanced epithelial ovarian cancer based on presence or absence of miliary disease. Following the Aust and Pils line of investigation (11, 13), our study represents the largest clinical investigation of these potential disease subtypes. Their work suggests that there exists an underlying biological difference between bulky implants and miliary disease, which could present distinct mechanisms of disease spread within the abdomen. Future studies of peritoneal implants should consider a range of genomics tests, ideally multiple sites from the same patient.

Highlights.

1. High-stage ovarian cancers commonly present with miliary disease.

2. Miliary disease may be underemphasized by size and location based staging systems.

3. Accounting for staging, cases with miliary spread are more likely to recur and have shorter survival.