Transcript Document

Cervical Cancer Case Study
Supervising Professor: Dr. P.D.M. Macdonald
Team Members: Christine Calzonetti, Simo
Goshev, Rongfang Gu, Shahidul
Mohammad Islam, Amanda
Lafontaine, Marcus Loreti, Maria
Porco, William Volterman, Qihao
Xie
-McMaster University-
Objectives:
•To determine which of the
documented variables are useful
for predicting recurrence of the
disease
•To evaluate the extent to which
tumor size, in particular, predicts
the recurrence of the disease
Graphical
Analysis
60
50
40
30
20
Age of Patient at time of Diagnosis (Years)
70
Boxplot of Age of Patient for Relapse Vs. Non-relapse Patients
Non-relapse
Relapse
Mean
Median
Non-relapse
42.08
40
Standard
deviation
11.04
Relapse
42.04
39
11.17
•The
majority of patients observed were between the ages of 35
and 50
•No
significant difference between relapse and non-relapse
patients
40
30
20
10
0
Depth of Tumor at Diagnosis (mm)
50
Boxplot of Tumor Depth for Relapse Vs. Non-relapse Patients
Non-relapse
Relapse
Mean
Median
Standard
deviation
Non-relapse
6.76
5
6.83
Relapse
7.71
11
10.11
•Similar means
•Dissimilar boxplots possibly due to outliers
•Missing values in the relapse group may have affected the outcome
40
20
0
Size of Tumor at Diagnosis (mm)
60
Boxplot of Tumor Size for Relapse Vs. Non-relapse Patients
Non-relapse
Non-relapse
Relapse
Relapse
Mean
Median
8.07
18.86
0
20
Standard
deviation
10.17
16.31
•A great disparity exists between the means and variability of relapse
and non-relapse patients
•Relapse patients had larger tumor sizes upon diagnosis, suggesting
that tumor size should be considered an important prognostic factor
Lymph Node Status for Non-Relapse Patients
Lymph Node Status for Relapse Patients
0
0
1
1
•The difference in pie charts indicates that there are
more cancerous cells found in the lymph nodes of
patients who relapsed
•The statistical significance is unclear
Cell Grade for Non-Relapse Patients
Cell Grade for Relapse Patients
1
2
1
0
0
2
3
3
•Relapse patients had a greater quantity of
cells deemed “worse”
Disease Status for Non-Relapse Patients
Disease Status for Relapse Patients
0
1
0
5
4
12
3
2
•Recorded at the time of follow up appointment (therefore cannot
be used as a diagnostic factor)
•Most non-relapse patients have no presence of disease at last
follow up appointment
•In relapse patients, approx. ½ died of disease, ¼ are alive with
disease, ¼ are alive with no evidence of disease
Results and
Conclusions
Survival Plot of Cervical Cancer Data
•Survival plot of data indicates that most relapses occur
during the first three years after surgery, it is highly unlikely
that relapse will occur after eight years
•The exponential curve deviated away from the survival curve
at the tail end due to the patients who will never relapse
Small (0-10mm)
Medium (11-30mm)
Large (30+mm)
Time
•Recurrence time for large group considerably
lower than medium
•Clear distinction between medium and small
•The patients in the different size groups had
noticeably different mean times to recur
Survival Analysis yielded the following results:
•Significant difference between medium and small
groups
•Significant difference between large and small groups
•Same results found using Weibull distribution in place
of exponential distribution
A survival analysis of the data on S-Plus where the exponential
distribution was assumed produced the following output:
Value
Std. Error
z
p
(Intercept)
9.275
0.128
72.21 0.00e+000
cutsize1
-0.552
0.139
-3.97 7.06e-005
cutsize2
-0.670
0.100
-6.67 2.48e-011
Regression Analysis yielded the following results:
A step-wise regression analysis of the data on S-Plus where the
exponential distribution was assumed produced the following
output: Initial variables:
Value
Std. Error
(Intercept)
11.0113
0.8091
13.6092 3.53e-042
cutsize1
-0.0615
0.1796
-0.3425 7.32e-001
cutsize2
-0.3278
0.1618
-2.0256 4.28e-002
lymph
-0.7694
0.4661
-1.6508 9.88e-002
depth
-0.0703
0.0146
-4.7977 1.61e-006
grad
-0.5229
0.2063
-2.5349 1.12e-002
age
0.0142
0.0145
0.9758 3.29e-001
rad
0.0245
0.2966
0.0827 9.34e-001
Final variables:
z
p
Value
Std. Error
z
p
(Intercept)
11.5989
0.5526
20.99 8.03e-098
cutsize1
-0.0660
0.1786
-0.37 7.12e-001
cutsize2
-0.3292
0.1609
-2.05 4.08e-002
lymph
-0.7735
0.3973
-1.95 5.16e-002
depth
-0.0666
0.0141
-4.71 2.46e-006
grad
-0.5378
0.2063
-2.61 9.15e-003
•Initial analysis showed that possible prognostic
factors were Size, Lymph Nodes, Tumor Depth and
Cell Grade
•Cox’s Proportional Hazard reaffirmed that Size,
Depth and Cell Grade were important diagnostic
factors, but Lymph Nodes are only significant at the
10% level