SEARCH

SEARCH BY CITATION

Keywords:

  • prognostic model;
  • early stage;
  • cervical cancer;
  • surgery

Abstract

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. CONFLICT OF INTEREST DISCLOSURES
  8. REFERENCES

BACKGROUND:

In the management of early stage cervical cancer, knowledge about the prognosis is critical. Although many factors have an impact on survival, their relative importance remains controversial. This study aims to develop a prognostic model for survival in early stage cervical cancer patients and to reconsider grounds for adjuvant treatment.

METHODS:

A multivariate Cox regression model was used to identify the prognostic weight of clinical and histological factors for disease-specific survival (DSS) in 710 consecutive patients who had surgery for early stage cervical cancer (FIGO [International Federation of Gynecology and Obstetrics] stage IA2-IIA). Prognostic scores were derived by converting the regression coefficients for each prognostic marker and used in a score chart. The discriminative capacity was expressed as the area under the curve (AUC) of the receiver operating characteristic.

RESULTS:

The 5-year DSS was 92%. Tumor diameter, histological type, lymph node metastasis, depth of stromal invasion, lymph vascular space invasion, and parametrial extension were independently associated with DSS and were included in a Cox regression model. This prognostic model, corrected for the 9% overfit shown by internal validation, showed a fair discriminative capacity (AUC, 0.73). The derived score chart predicting 5-year DSS showed a good discriminative capacity (AUC, 0.85).

CONCLUSIONS:

In patients with early stage cervical cancer, DSS can be predicted with a statistical model. Models, such as that presented here, should be used in clinical trials on the effects of adjuvant treatments in high-risk early cervical cancer patients, both to stratify and to include patients. Cancer 2011. © 2010 American Cancer Society.

Early stage cervical cancer has a relatively favorable prognosis and can in most patients be controlled by radical surgery or (chemo-)radiotherapy.1, 2 Pelvic lymph node metastases are the most important risk factor in surgically treated patients for recurrence or failure to survive. Adjuvant therapy is therefore frequently advised for this group of patients.3, 4 However, approximately 50% of recurrences occur in patients without lymph node metastases.5, 6 Other pathological risk factors, including parametrial extension, tumor-positive surgical margins, tumor size, lymph vascular space invasion, and depth of invasion have been identified.

Traditionally, adjuvant radiotherapy after radical surgery has been used for patients with positive lymph nodes, parametrial involvement, and/or positive margins, although no randomized controlled trial (RCT) has been performed on this subject. Recently, concurrent chemotherapy and radiotherapy have been introduced for patients with positive lymph nodes on the basis of a RCT.7-9

Currently, even for patients with negative lymph nodes but poor prognostic tumor parameters (such as deep stromal invasion, vascular space invasion, and large tumor diameter) some advocate adjuvant radiotherapy,10-12 whereas others refute this policy.13

However, there is no consensus on the relative importance of the prognostic clinical and pathological factors for the individual patient.13-15

Comparison of the outcomes in various reported series is difficult, because prognostic factors may differ in importance, depending on patient selection, the definition of prognostic factors, the radicality of the surgery, and the differences in guidelines for adjuvant treatment.16, 17 Moreover, disease-specific survival (DSS) is more relevant than the occurrence of recurrent disease, although most of the variables are identified to predict recurrence of disease.

The aim of this study is to develop a prognostic model to predict DSS in patients treated primarily with radical surgery for early stage cervical cancer. Moreover, high-risk patients who might benefit from additional treatment (such as chemotherapy, biological therapy, or hyperthermia) are identified.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. CONFLICT OF INTEREST DISCLOSURES
  8. REFERENCES

Patients

Consecutive patients who had surgery for early stage cervical cancer (International Federation of Gynecology and Obstetrics [FIGO] stage IA2-IIA) between 1982 and 2007 in the Academic Medical Center (Amsterdam, the Netherlands) were included. Staging occurred according to the guidelines of the FIGO system; these did not change during the inclusion period. Patients referred from foreign countries were excluded, because an adequate follow-up was not feasible. Only patients with squamous, adenosquamous, or adenocarcinoma were included. All patients were treated with a radical hysterectomy and pelvic lymph node dissection according to Wertheim-Okabayashi; this type of surgery has been described by Samlal et al.5

The following data were retrieved from the database, patient files, and pathology reports: age, FIGO stage, vaginal wall extension, state of surgical margins, cell type, differentiation grade, clinical tumor size, lymph node involvement, number of positive lymph nodes, parametrial involvement, depth of tumor invasion (in millimeters and in fractional thirds), lymph vascular space invasion, adjuvant treatment, and follow-up status. If there was no residual tumor after conization, we could not assess depth of invasion in thirds. As all of these tumors had a diameter <2 cm and infiltration of <10 mm in the stroma, we included these tumors in the group with <⅓ depth of infiltration. Pathology until 1992 was retrospectively reviewed by an expert gynecologic pathologist, whereas pathology results from 1992 onward were taken from the tumor board notes.

Indications for adjuvant radiotherapy were lymph node metastasis, parametrial infiltration, tumor-positive surgical margins, or spill during surgery. From 2001, patients with >1 risk factor were offered concurrent chemotherapy and radiotherapy. Risk factors in this respect were adenocarcinoma, lymph node metastases, and parametrial infiltration. All patients were attending the outpatient clinic for follow-up visits. Routine follow-up was continued up to 5 years. At each visit, patients were asked about complaints, and gynecological examination was performed. When there was a clinical suspicion of recurrent disease, diagnostic tests such as magnetic resonance imaging, computed tomography, serum tumor marker(s), and/or biopsy were performed.

The primary endpoint in this study was DSS, defined as the time to death related to cervical cancer.

Statistical Analysis

The aim of the analysis was to study the associations between the prognostic variables and the DSS. Missing data of the prognostic variables were imputed (filled in), because deleting them would lead to a loss of statistical power and potentially biased results.18-20 We generated a single imputed data set, using the first step of the aRegImpute multiple imputation function in Splus6.0.21 This is an efficient implementation of Bayesian multiple imputation.22

The linearity of the continuous variables tumor diameter, depth of tumor invasion, and age was assessed using spline functions.23 Nonlinear associations were redefined, based on these spline functions. Survival rates were estimated by the Kaplan-Meier method.24 Patients were censored at the date of last visit or at the time of death not related to the cervical cancer. Univariate and multivariate regression analyses were performed. For the multivariate Cox proportional hazard regression analysis, a stepwise backward selection procedure was used.25, 26 As the incorrect exclusion of a factor would be more deleterious than including too many factors, all prognostic variables with a significance level up to 30% were included in the univariate analysis.27

Internal validation was performed by the bootstrap method, in which new data sets are created by random drawing from the sample with replacement.26, 28, 29 In each of these new data sets (n = 200), multivariate regression analyses were performed. By analyzing the difference of the prognostic models, a shrinkage factor was calculated to estimate the overfit of the created model; the model was corrected accordingly.26, 30

Calibration of the model was assessed by comparing the predicted DSS and the observed DSS at 5-year follow-up. The point estimates of the predicted and observed 5-year survival in the subgroups were graphically depicted. The calibration plot allows for a visual assessment of the performance of the model. In case of a perfect calibration, all predictions and observations would be located on the line of equality (X = Y), and the slope would be 1.

To evaluate the discriminative capacity of the prognostic model, the area under the receiver operating characteristic (ROC) curve was calculated.31 We used the state of disease at the last date of follow-up or the time of disease-related death, in relation to the predicted survival at this time. Sensitivity was defined as the fraction of correctly predicted surviving patients, whereas specificity was defined as the fraction of patients who died of disease that was predicted correctly. Next, the discriminative capacity of the prognostic model was calculated at 5-year follow-up. For this, we used the state of disease of the patients at 4.5 to 5 years follow-up in relation to the predicted survival at 5 years.

Score Chart

The model produced is the most reliable for predicting DSS at different times of follow-up. However, because this model may be too complicated for daily practice, we converted the parameters from the logistic Cox model to a simple additive score chart, which predicts the DSS at 5-year follow-up.28 First, the continuous factor tumor size was divided into subgroups based on the spline function.23 The continuous factor “depth of tumor invasion” was categorized according to conventionally used fractional thirds. Subsequently, the multivariate Cox proportional hazard regression analysis was repeated. The regression coefficients were corrected for the shrinkage factor and divided through the smallest coefficient to obtain round numbers in the score chart.26, 30

Subsequently, the predicted DSS was calculated from the model and the observed DSS. We calibrated this clinical prediction rule based on the score chart and evaluated the discriminative capacity by calculating the area under the ROC curve. A subdivision into 4 groups was made; a low-risk group with a predicted DSS of >95%, an intermediate-risk group with a predicted DSS from 85% to 95%, a high-risk group with a predicted DSS from 70% to 85%, and a very high-risk group with a predicted DSS of <70%.

Performance of the Score Chart in Our Patients

Although the score chart has only been validated internally, we looked at the performance of the score chart in our patients to show the clinical relevance. In the risk groups, the adjuvant therapy that was given was analyzed in relation to recurrent disease.

RESULTS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. CONFLICT OF INTEREST DISCLOSURES
  8. REFERENCES

Patients

From 1982 to 2007, 768 patients surgically treated for early stage cervical cancer were registered in the database. In total, 45 patients were excluded, because follow-up was not feasible, and 13 patients with a small cell carcinoma or a sarcoma were excluded because tumor behavior and treatment strategy differed substantially. Therefore, the study group consisted of 710 patients. As each patient had 11 variables, there were 7810 data points. In total, 253 (3.2%) data points were missing and subsequently imputed. Table 1 gives a summary of the patient's characteristics. The mean age was 41 (range, 16-77) years. The mean follow-up was 62 (range, 9-127) months. Most patients had clinical FIGO stage IB (88%) and squamous cell carcinoma (74%). Three patients with stage IA2 were treated with radical surgery: 2 because of lymph vascular space invasion and 1 because the surgical margins of the exconization were dubious. Lymph node metastases were present in 19% of patients, whereas 17% had parametrial invasion, and 40% had lymph vascular space invasion. A total of 217 (31%) patients received adjuvant treatment, and 1 patient received neoadjuvant chemotherapy before surgery.

Table 1. Characteristics of the Study Population
CharacteristicsBefore ImputationAfter Imputation
No. of Patients%No. of Patients%
  • FIGO indicates International Federation of Gynecology and Obstetrics; SCC, squamous cell carcinoma; AC, adenocarcinoma; ASC, adenosquamous cell carcinoma; LNM, lymph node metastasis; LVSI, lymph vascular space invasion; RT, radiotherapy; CT, chemotherapy.

  • a

    Value shown is the 95% confidence interval.

  • b

    Tumor down to/close to the surgical margins.

Age, median y4116-77aUnchanged 
Follow-up, median mo629-127aUnchanged 
FIGO stage
 IA231Unchanged 
 IB153475  
 IB29613  
 IIA7711  
Histology
 SCC52774Unchanged 
 AC15522  
 ASC304  
Differentiation grade
 Well436507
 Moderate2613731444
 Poor3054334649
 Unknown10114  
Tumor diameter, mm
 Median308-56a308-60a
 Unknown537  
Depth of invasion, mm
 Median93-20a93-20a
 Unknown7711  
Stromal invasion
 <1/32383424535
 1/3-2/31432015121
 >2/33074331444
 Unknown223  
Vaginal wall extension
 Yes7711Unchanged 
 No63389  
Parametrial extension
 Yes12317Unchanged 
 No58783  
LNM
 Yes13219Unchanged 
 No57881  
LVSI
 Yes28540Unchanged 
 No42560  
Surgical margins/spill
 Free68697Unchanged 
 Dysplasia71  
 Not freeb172  
 Adjuvant treatment,  RT (±CT)21731Unchanged 

Statistical Analysis

Overall, the 5-year DSS was 92%. The Kaplan Meier curves for DSS stratified by FIGO stage are shown in Figure 1. For stage IB1, IB2, and IIA, the 5-year DSS rates were 94%, 85%, and 89%, respectively.

thumbnail image

Figure 1. Kaplan-Meier curves for disease-specific survival are shown stratified by International Federation of Gynecology and Obstetrics (FIGO) stage.

Download figure to PowerPoint

Both age and tumor size were linearly associated with DSS. For depth of invasion, a linear association was only found for values up to 10 mm, therefore all measurements deeper than 10 mm were treated as equivalent to 10 mm (Fig. 2).

thumbnail image

Figure 2. Spline functions are shown. (A) The association between the continuous variable depth of invasion and disease-specific survival is shown. (B) The association between the continuous variable tumor diameter and disease-specific survival is shown.

Download figure to PowerPoint

Table 2 shows the results of the univariate and multivariate analyses. Tumor size, histological type, presence of lymph node metastasis, depth of stromal invasion (in millimeters), lymph vascular space invasion, and parametrial extension were all independently associated with DSS.

Table 2. Results of the Univariate and Multivariate Regression Analysis
VariableUnivariate AnalysisMultivariate Analysis
HR95% CIPβaHR95% CIP
  • HR indicates hazard ratio; CI, confidence interval; β, regression coefficient; SCC, squamous cell carcinoma; Ref, reference; AC, adenocarcinoma; ASC, adenosquamous cell carcinoma; LNM, lymph node metastasis; LVSI, lymph vascular space invasion; DSS, disease-specific survival; TD, tumor diameter; DI, depth of invasion.

  • equation image.

  • a

    The 5-year disease-specific survival (DSS) can be calculated from the multivariate model with the formula:

  • equation image

  • b

    Per millimeter up to 10 mm linear, 10 mm or more equals 10.

Age, y1.00.96-1.02.40    
Histological type
 SCCRef  Ref   
 AC2.71.5-4.9.0011.64.82.6-9.1<.001
 ASC2.81.0-8.2.050.461.60.55-4.6.40
Differentiation grade
 WellRef      
 Moderate1.60.39-7.1.49    
 Poor2.00.49-8.9.33    
Tumor diameter, mm1.031.02-1.05<.0010.021.021.0-1.04.05
Depth of invasion, mmb1.61.3-2.0<.0010.241.31.01-1.6.04
Stromal invasion
 <1/3Ref      
 1/3-2/33.40.89-13.07    
 >2/39.93.1-32<.001    
Vaginal wall extension macroscopic1.50.69-3.1.31    
Parametrial extension present5.63.2-9.6<.0011.083.01.6-5.4<.001
LNM present7.34.1-13<.0011.23.31.8-6.1<.001
LVSI present3.41.9-6.1<.0010.852.31.2-4.5.01
Surgical margins/spill positive0.930.56-1.5.77    

Internal validation by bootstrapping produced a shrinkage factor of 0.91, which corresponds to 9% overfit of the model. The DSS at time × tx (after correction by the shrinkage factor) can be calculated from the multivariate model with the formula:

  • equation image

where tx is time at point x, PMI is parametrial invasion, TD is tumor diameter, AC is adenocarcinoma, ASC is adenosquamous cell carcinoma, LNM is lymph node metastasis, DI is depth of invasion, and LVSI is lymph vascular space invasion.

The formula to predict the 5-year DSS is:

  • equation image

The association between the estimated and observed DSS was good (data not published). The prognostic model had an area under the ROC curve of 0.73 (95% confidence interval [CI], 0.66-0.81). The area under the ROC curve for the prediction of the 5-year DSS was 0.85 (95% CI, 0.79-0.92).

Score Chart

For the score chart, the tumor diameter was divided into 3 subgroups based on the spline function (<20 mm, 20-60 mm, and ≥60 mm). For the depth of invasion, we used the fractional thirds. Next, we made a new model for the 5-year DSS and corrected the regression coefficients (β) using the shrinkage factor (data not published). The association between the predicted DSS, as calculated from our score chart, and the observed DSS was good, meaning that the 95% error bars of the observed DSS crossed the line of equality. The area under the ROC curve was 0.85 (95% CI, 0.79-0.92), indicating a good discriminative capacity (Fig. 3). Table 3 shows the score chart for estimation of the probability of 5-year DSS after surgically based treatment. The sum score for a patient can range from 0 (best prognosis) to 19 (worst prognosis). The relation between the sum score and the estimated 5-year DSS is depicted in Figure 4. Table 4 shows the grouping of risk into 4 risk categories with the correlated predicted DSS.

thumbnail image

Figure 3. The receiver operating characteristic curve of the score chart for the prediction of disease-specific survival at 5 years of follow-up is shown. Area under the curve is 0.85 (95% confidence interval, 0.78-0.92).

Download figure to PowerPoint

thumbnail image

Figure 4. Probability of 5-year disease-specific survival (DSS) is shown by prognostic sum score.

Download figure to PowerPoint

Table 3. Score Chart for Estimation of the Probability of Disease-Specific Survival
CharacteristicScoresSum Score
  1. AC indicates adenocarcinoma; ASC, adenosquamous cell carcinoma; LNM, lymph node metastasis; LVSI, lymph vascular space invasion.

Histology
 AC4
 ASC1
Tumor diameter, mm
 20-601
 ≥604
LNM present3
Parametrial extension present3
LVSI present2
Depth of invasion
 1/3-2/31
 ≥2/33
Sum score (add all) maximum19…….
Table 4. Grouping of Risk and Predicted DSS With the Given Therapy With Rate and Location of Recurrent Disease in Our (AMC) Patient Group Based on the Prognostic Risk Group
Risk GroupSum ScorePredicted DSSAMC Patients
No.Adjuvant TherapyNo.5-Year DSSRecurrences
Total%LocalDistantBothUnknown
  • DSS indicates disease-specific survival; AMC, Academic Medical Center, Amsterdam; RT, radiotherapy; CRT, chemoradiation.

  • a

    Including 1 patient treated with neoadjuvant chemotherapy.

Low0-6>95%475Nonea4399817410421
RT33100-0----
CRT3100-0----
Intermediate7-985%-95%123None44977164111
RT7294710241-
CRT783114-1--
High10-1265%-85%86None106544021-1
RT4876132766-1
CRT2878414-4--
Very high13-19<65%26None00-0----
RT15311067352-
CRT1131545-32-

Performance of the Score Chart in Our Patients

Table 4 shows the adjuvant therapy in relation to the rate and location of recurrences in our population.

In the low-risk group, 8% received adjuvant treatment; none of these patients showed recurrent disease. Of the patients without adjuvant treatment, 4% developed a recurrence; 2% were solely local. The 5-year DSS in this group without adjuvant treatment was 98%.

In the intermediate-risk group, 44 (36%) patients received no adjuvant treatment. Seven (16%) patients had recurrent disease; the 5-year DSS was 97%.

Of the 12% of patients in the high-risk group who did not receive adjuvant treatment, 4 (40%) patients had recurrent disease, compared with 20% in the group receiving adjuvant treatment. The 5-year DSS was higher in the patients receiving adjuvant treatment; that is, 76% and 78% in the radiotherapy and chemoradiation groups, respectively, compared with 65% in patients without adjuvant treatment.

All patients in the very high-risk group received adjuvant treatment; nevertheless, the recurrence rate was high (58%). The 5-year DSS was not reduced by adding chemotherapy to the radiotherapy in this group.

DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. CONFLICT OF INTEREST DISCLOSURES
  8. REFERENCES

This study presents a prognostic model for surgically treated early stage cervical cancer patients predicting the DSS. Overall, the 5-year DSS was 92%. Tumor diameter, histological type, lymph node metastasis, depth of stromal invasion, lymph vascular space invasion, and parametrial extension were independently associated with DSS. All these factors were included in the prognostic model. The performance of the derived clinical score chart, predicting 5-year DSS, was good, with an accurate calibration and a good discriminative capacity.

The present study has some limitations. First, because cervical cancer has a low incidence, the time period in which the cohort of patients was collected was 25 years. However, during the study period, the surgical technique and indications for adjuvant treatment were uniform at our oncology center, except for the described change of adding concomitant chemotherapy to the radiotherapy. Second, in the ideal situation the total population used to develop the prognostic model should receive the same treatment. In general, adjuvant treatment will favor survival and, therefore, a subgroup of our patients received adjuvant treatment.9-11 One can argue that some of the prognostic factors used in the model were already used to indicate adjuvant therapy and may have had more influence on survival if no adjuvant therapy was applied. However, all the factors in the model remained independent predictors of DSS, despite extensive surgery and adjuvant radiotherapy. Third, early stage cervical cancer has a good prognosis, that is, 92% DSS in our group. Because of the low number of events, the number of patients had to be large to obtain reliable results.

A strength of this study is that factors in the prognostic model were included according to their weight given in the multivariate analysis. This model makes it possible to give a better individual prediction of the risk of disease-specific death after radical surgery, with or without adjuvant treatment, at different times of follow-up.

The subtracted clinical score chart still uses all the prognostic factors. However, DSS is predicted only at 5-year follow-up.

One has to bear in mind that this model was constructed in a group of patients treated in a single hospital. External validation is important, because internal validation is known to overestimate the performance of models.32-34 Therefore, validation in an independent group of patients should be performed. However, because it was the best alternative available, we used internal validation, which in other populations showed an estimated overfit of 9%, and corrected the model based on this overfit.

In the Dutch guidelines, adjuvant radiotherapy is advocated in case of lymph node metastasis, parametrial involvement, or positive surgical margins (www.oncoline.nl). In the present study, all patients with these conditions received radiotherapy. Positive surgical margins were not associated with survival in the present study, probably because all these patients received adjuvant treatment, and this feature was seen in only 2% of our patients. One of the most commonly used prognostic models in early stage cervical cancer is the Delgado model, later adapted by Sedlis et al.11, 15, 16 These models were developed only for patients with squamous cell type of cervical cancer. Our model shows that adenocarcinoma has an important impact on survival (hazard ratio, 4.4). Rotman et al reported that adjuvant radiotherapy appeared to be particularly beneficial for patients with adenocarcinoma or adenosquamous carcinoma.10 Peters et al also showed a poorer prognosis for adenocarcinomas or adenosquamous carcinomas and suggested that these patients might benefit more from adjuvant treatment than patients with squamous cell carcinoma.9, 10 However, neither the Sedlis criteria nor the Dutch guidelines used this prognostic factor to indicate adjuvant treatment.

Addition of chemotherapy to radiotherapy enhances survival in cervical cancer patients.7-9 In our population, the overall DSS is good with the current (conservative) treatment strategies (92%). It remains unclear which of the early stage cervical cancer patients will benefit from adjuvant chemotherapy concomitant with radiotherapy.35

The percentage of recurrences in the intermediate-risk group did not differ between patients treated with adjuvant radiotherapy and patients with concomitant chemotherapy. The significance of the lower 5-year DSS in the latter group is not clear, because this involved only 7 patients. In the high-risk and very high-risk groups, the recurrence rate was lower in the patients treated with adjuvant chemoradiation compared with adjuvant radiotherapy alone, and this difference was even greater compared with no adjuvant treatment at all. The gain seems to be a better local control of the disease. However, no benefit of concomitant chemotherapy was seen in the 5-year DSS.

Until further clinical studies have shown a better approach, we recommend adjuvant treatment for patients in the high-risk and very high-risk groups (sum score ≥10). However, new (adjuvant) treatment strategies are urgently needed, particularly treatments that lower the risk of distant recurrences.

On the basis of the clinical score chart, recommending a change in treatment strategies for the low-risk and intermediate-risk groups remains difficult, because it is unknown whether the good prognosis is in fact because of the treatment (radical surgery with or without adjuvant [chemo-]radiotherapy). Clinical trials are needed to establish whether reducing the indications for adjuvant treatment in cervical cancer will have some impact on survival.

In conclusion, tumor diameter, histological type, presence of lymph node metastasis, depth of stromal invasion, lymph vascular space invasion, and parametrial extension were independently associated with DSS. On the basis of these factors, we developed a simple prognostic model that accurately identifies risk groups. Models such as that presented here should be used in RCTs on the effects of adjuvant treatments in high-risk early cervical cancer patients, both to stratify and to select patients.

Acknowledgements

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. CONFLICT OF INTEREST DISCLOSURES
  8. REFERENCES

We thank Nan van Geloven for her comments and advice on the statistical procedures.

REFERENCES

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. CONFLICT OF INTEREST DISCLOSURES
  8. REFERENCES
  • 1
    Eifel PJ, Morris M, Wharton JT, Oswald MJ. The influence of tumor size and morphology on the outcome of patients with FIGO stage IB squamous cell carcinoma of the uterine cervix. Int J Radiat Oncol Biol Phys. 1994; 29: 9-16.
  • 2
    Landoni F, Maneo A, Colombo A, et al. Randomised study of radical surgery versus radiotherapy for stage Ib-IIa cervical cancer. Lancet. 1997; 350: 535-540.
  • 3
    Sevin BU, Nadji M, Lampe B, et al. Prognostic factors of early stage cervical cancer treated by radical hysterectomy. Cancer. 1995; 76: 1978-1986.
  • 4
    Soisson AP, Soper JT, Clarke-Pearson DL, Berchuck A, Montana G, Creasman WT. Adjuvant radiotherapy following radical hysterectomy for patients with stage IB and IIA cervical cancer. Gynecol Oncol. 1990; 37: 390-395.
  • 5
    Samlal RA, van der Velden J, Ketting BW, et al. Disease free interval and recurrence pattern after the Okabayashi variant of Wertheim's radical hysterctomy for stage IB and IIA cervical carcinoma. Int J Gynecol Cancer. 1996; 6: 120-127.
  • 6
    Smiley LM, Burke TW, Silva EG, Morris M, Gershenson DM, Wharton JT. Prognostic factors in stage IB squamous cervical cancer patients with low risk for recurrence. Obstet Gynecol. 1991; 77: 271-275.
  • 7
    Keys HM, Bundy BN, Stehman FB, et al. Cisplatin, radiation, and adjuvant hysterectomy compared with radiation and adjuvant hysterectomy for bulky stage IB cervical carcinoma. N Engl J Med. 1999; 340: 1154-1161.
  • 8
    Morris M, Eifel PJ, Lu J, et al. Pelvic radiation with concurrent chemotherapy compared with pelvic and para-aortic radiation for high-risk cervical cancer. N Engl J Med. 1999; 340: 1137-1143.
  • 9
    Peters WA III, Liu PY, Barrett RJ, et al. Concurrent chemotherapy and pelvic radiation therapy compared with pelvic radiation therapy alone as adjuvant therapy after radical surgery in high-risk early-stage cancer of the cervix. J Clin Oncol. 2000; 18: 1606-1613.
  • 10
    Rotman M, Sedlis A, Piedmonte MR, et al. A phase III randomized trial of postoperative pelvic irradiation in stage IB cervical carcinoma with poor prognostic features: follow-up of a gynecologic oncology group study. Int J Radiat Oncol Biol Phys. 2006; 65: 169-176.
  • 11
    Sedlis A, Bundy BN, Rotman MZ, Lentz SS, Muderspach LI, Zaino RJ. A randomized trial of pelvic radiation therapy versus no further therapy in selected patients with stage IB carcinoma of the cervix after radical hysterectomy and pelvic lymphadenectomy: a Gynecologic Oncology Group Study. Gynecol Oncol. 1999; 73: 177-183.
  • 12
    Grigsby PW. Cervical cancer: combined modality therapy. Cancer J. 2001; 7( suppl 1): S47-S50.
  • 13
    van der Velden J, Samlal R, Schilthuis MS, Gonzalez DG, ten Kate FJ, Lammes FB. A limited role for adjuvant radiotherapy after the Wertheim/Okabayashi radical hysterectomy for cervical cancer confined to the cervix. Gynecol Oncol. 1999; 75: 233-237.
  • 14
    Rotman M, Sedlis A, Piedmonte MR, et al. A phase III randomized trial of postoperative pelvic irradiation in stage IB cervical carcinoma with poor prognostic features: follow-up of a gynecologic oncology group study. Int J Radiat Oncol Biol Phys. 2006; 65: 169-176.
  • 15
    Biewenga P, van der Velden J, Mol BW, et al. Validation of existing prognostic models in patients with early-stage cervical cancer. Gynecol Oncol. 2009; 115: 277-284.
  • 16
    Delgado G, Bundy B, Zaino R, Sevin BU, Creasman WT, Major F. Prospective surgical-pathological study of disease-free interval in patients with stage IB squamous cell carcinoma of the cervix: a Gynecologic Oncology Group study. Gynecol Oncol. 1990; 38: 352-357.
  • 17
    Sevin BU, Lu Y, Bloch DA, Nadji M, Koechli OR, Averette HE. Surgically defined prognostic parameters in patients with early cervical carcinoma. A multivariate survival tree analysis. Cancer. 1996; 78: 1438-1446.
  • 18
    Little RJ, Rubin DB. Causal effects in clinical and epidemiological studies via potential outcomes: concepts and analytical approaches. Annu Rev Public Health. 2000; 21: 121-145.
  • 19
    Schafer JL. Analysis of Incomplete Multivariate Data. London, UK: Chapman & Hall; 1997.
  • 20
    Moons KG, Donders RA, Stijnen T, Harrell FE Jr. Using the outcome for imputation of missing predictor values was preferred. J Clin Epidemiol. 2006; 59: 1092-1101.
  • 21
    Harrell FE.TheHmisclibrary.Available at: http://biostat.mc.vanderbilt.edu/wiki/Main/Hmisc. 2008 Accessed on…
  • 22
    Schafer JL, Graham JW. Missing data: our view of the state of the art. Psychol Methods. 2002; 7: 147-177.
  • 23
    Harrell FE Jr, Lee KL, Pollock BG. Regression models in clinical studies: determining relationships between predictors and response. J Natl Cancer Inst. 1988; 80: 1198-1202.
  • 24
    Kaplan EL, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc. 1958; 53: 457-481.
  • 25
    Cox DR. Regression models and life-tables. J R Stat Soc. 1972; 34: 187-220.
  • 26
    Harrell FE. Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. New York, NY: Springer; 2001.
  • 27
    Steyerberg EW, Eijkemans MJ, Habbema JD. Stepwise selection in small data sets: a simulation study of bias in logistic regression analysis. J Clin Epidemiol. 1999; 52: 935-942.
  • 28
    Steyerberg EW. Clincial Prediction Models. A Practical Approach to Development, Validation, and Updating. New York, NY: Springer; 2009.
  • 29
    Steyerberg EW, Bleeker SE, Moll HA, Grobbee DE, Moons KG. Internal and external validation of predictive models: a simulation study of bias and precision in small samples. J Clin Epidemiol. 2003; 56: 441-447.
  • 30
    Van Houwelingen JC, Le CS. Predictive value of statistical models. Stat Med. 1990; 9: 1303-1325.
  • 31
    Heagerty PJ, Lumley T, Pepe MS. Time-dependent ROC curves for censored survival data and a diagnostic marker. Biometrics. 2000; 56: 337-344.
  • 32
    Bleeker SE, Moll HA, Steyerberg EW, et al. External validation is necessary in prediction research: a clinical example. J Clin Epidemiol. 2003; 56: 826-832.
  • 33
    Stolwijk AM, Zielhuis GA, Hamilton CJ, et al. Prognostic models for the probability of achieving an ongoing pregnancy after in-vitro fertilization and the importance of testing their predictive value. Hum Reprod. 1996; 11: 2298-2303.
  • 34
    Wasson JH, Sox HC, Neff RK, Goldman L. Clinical prediction rules. Applications and methodological standards. N Engl J Med. 1985; 313: 793-799.
  • 35
    Monk BJ, Wang J, Im S, et al. Rethinking the use of radiation and chemotherapy after radical hysterectomy: a clinical-pathologic analysis of a Gynecologic Oncology Group/Southwest Oncology Group/Radiation Therapy Oncology Group trial. Gynecol Oncol. 2005; 96: 721-728.