Cholangiocarcinoma.org Statistics (CONTAINS SURVIVAL STATISTICS)

Hi Eli,

The R program can do multvariate analysis, but I don’t think it is within it’s survival package. The survival package is designed to combine times to actual deaths and times to when patients leave the study while still alive, referred to as right-censored data. I have done multivariate analysis before using Lotus, but right now, I don’t have a feel on how I would combine actual deaths and right-censored data within a mutivariate analysis.

I was surprised by my results, because I haven’t seen any cc survival studies involving sex and age. I did notice that the Cisplatin plus Gemcitabine versus Gemcitabine for Biliary Tract Cancer Phase III Clinical Trial balanced age and sex in their studies. Median age for Gem was 63.2 yrs. and 63.9 yrs for GemCis. Also, there were 108 women and 98 men for Gem, while there were 108 women and 96 men for GemCis. So it appears they were trying to negate any possible sex and age influence on the results. The survival curves shown in Fig. 2 of their paper were only separated by a few months, which was much closer than the curves I show in the two previous plots.

I have a lot more men than women in my data mostlly because about four times as many wives post about their husbands to these boards than husbands post about their wives. Daughters and sons posts about evenly for their parents, but daughters outnumber sons in the posts by over four to one.

Somewhere in the future, I will show survival curves for more than one variable, so I could include sex, age, and resection status in one survival plot which would allow comparisons. The problem is that as I include more variables, the sample size for each condition decreases and p increases above 0.05, so I lose statistical significance. I correct this by compiling more data from the posts, but this is very time consuming. Also, I spend a lot of time looking online for possible deaths of members who no longer post because the survival analyses starts to lose accuracy if too many right-adjusted data points are used.

Bruce

[4/25/13 – Eli, since I posted this I learned that the Cox Regression Model is used in survival analysis to handle the right-censored data for multivariate analyisis. The R Survival Package includes Cox regression analysis and I will eventually use it to do the analysis you suggested.]

The following plot depicts survival as a function of a patient’s age when his/her cholangiocarcinoma tumor is first detected. This is from a sample of 218 patients from my Cholangiocarcinoma.org database. The median age of this sample is 58 years and the plot contains survival data for patients who are less than 58 years old and for those who are age 58 and older. The data had a chisq of 9.5 , 1 degree of freedom, and a p=0.00208, indicating a high confidence level that the data is significant.

Combined with the plot for survival as a function of age, it appears that younger women fare the best and older men the worse.

I also came up with a survival plot comparing patients whose tumors were detected prior to 1/1/2007 vs. those who were detected after this date, but p was much greater than 0.05, indicating I need to compile more data for that plot to be statistically significant.

Bruce

OK, I had to read more about getting the plot from R onto this discussion topic.
Bruce

To date, I have compiled data for 347 members, which represents about 14.3% of this organization’s membership. Of these 347 members I have data for 251 cc patients for which I have a detection date, the date on which the patient was diagnosed with a tumor which was later identified as cc by additonal tests, pathology reports, etc. Many members post the dates as only a month and a year, and for these I estimate the actual date as the 15th of the month. Of these 251 patients, 146 have died (94 men and 52 women) and 105 (53 men and 52 women) are either still living or have died and I am not aware of their deaths. In survival studies, these 105 patients are treated as right-censored failures. They represent patients who drop out of studies. In those studies these patients and their times within the study are included in the analyses. In my study, I calculate these times from the last posts made by members in which the patient is still alive or from other sources such as blogs.

In order to process this data, I have downloaded the free R Statistical Program, which has a Survival Package and is used by the professionals for doing Kaplan-Meier estimates, log-rank tests, confidence intervals, Cox regression, and all the good stuff that the professionals who write the medical papers utilize.

My first try with this program was to calculate survivals for these 251 patients and compare men and women.

I used the Kaplan-Meier estimate and log-rank test, and found that the median survival time for men was 1.43 years and for women, 2.47 years. The test had a chisq of 7.5 , 1 degree of freedom, and a p=0.00603, indicating a high confidence level that the data is significant.

The Kaplan-Meier survival curves for men and women tracked together from 100% down to about 60%, after which the men’s curve continued its steep slope, while the slope of the women’s curve decreased, resulting in the greater median survival. At 30% survival, women were 1.67 years greater than men.

Bruce

Marion – I agree that more patients will improve the statistical significance of the results I presented, but the results are still worth presenting. Statistical significance will improve as I collect more data from the discussion boards and enter it into my spreadsheet.

50 patients were used in my analysis and were divided into two subsets, 33 from the USA and 17 not from the USA.

For comparison, the following article, published in the American Journal of Surgery, discusses a study with 31 patients and the authors were still able to make statments about survival for subsets of the 31 patients, those with R0 resection, negative lymph nodes, a solitary tumor, a width of resection margin greater than 3 mm, and stages III and IV patients who received chemotherapy. I don’t have the complete article, but I bet some of these subsets had less than 17 patients.

http://www.ncbi.nlm.nih.gov/pubmed/15720985

Bruce

Based on actual death data, the median survival time after diagnosis of cholangiocarcinoma for 17 patients living outside the United States is 0.64 years (0.10 – 3.37), whereas for 33 patients living inside the United States, it is double that value, or 1.28 years (0.12 – 5.45).

The countries in my database which are included as outside the USA are Australia, Belgium, Canada, Denmark, France, Greece, India, Mexico, and the United Kingdom.

Bruce Baird

So far I have death dates for 57 cc patients. Of these I was able to determine that 22 had resection surgery and 30 did not. The median survival was 2.27 years (0.63 yr. – 7.80 yr.) for those having resection and 0.98 year (0.02 yr – 2.78 yr.) for those who did not.

Considering that those who could not have surgery probably had more advanced cc than those who could, I looked at staging. Seven of those who had no surgery were classified stage IV and one was classified IIIC to IV. This latter case was diagnosed in Aug. 2011, when the 7th edition of AJCC Cancer Staging Manual was in effect. The patient had IHCC and there is no stage IIIC or IV for IHCC, only I, II, III, IVA and IVB. The surgeon evidently staged the patient based on the 6th edition of the Staging Manual which was out of effect on 1/1/2010.

Staging was reported for three patients who had surgery. One was stage I, one stage IIA, and one stage IV, enforcing the expected outcome that those who could not have surgery had more advanced cases of cc.

I will eventually calculate the stage for those patients who have provided the required TNM data but no stage so that the impact of surgery on advanced cc cases can be estimated.