вход по аккаунту



код для вставкиСкачать
The Relative Accuracy of the Clinical Estimation
of the Duration of Life for Patients with End
of Life Cancer
Antonio Viganò, M.D.1
Marlene Dorgan, M.L.S.2
Eduardo Bruera, M.D.1
Maria E. Suarez-Almazor,
BACKGROUND. Although the prediction of the duration of life of patients with end
Division of Palliative Care Medicine, Grey Nuns
Hospital, Edmonton, Alberta, Canada.
J. W. Scott Health Sciences Library, University of
Alberta, Edmonton, Alberta, Canada.
Department of Public Health Sciences, University
of Alberta, Edmonton, Alberta, Canada.
of life cancer most often relies on the clinical estimation of survival (CES) made by
the treating physician, the accuracy and practical value of CES remains controversial.
METHODS. The authors prospectively evaluated the accuracy of CES in an inception
and population-based cohort of 233 cancer patients who were seen at the onset of
their terminal phase. They also systematically reviewed the literature on CES in
advanced or end-stage cancer patients in MEDLINE, CANCERLIT, and EMBASE
data bases, using two search strategies developed by a research librarian.
RESULTS. CES had low sensitivity in detecting patients who died within shorter
time frames (#2 months), and a tendency to overestimate survival was noted. A
moderate correlation was observed between actual survival (AS) and CES (Pearson
correlation coefficient 5 0.47, intraclass correlation coefficient 5 0.46, weighted
kappa coefficient 5 0.42).
CONCLUSIONS. Treating physicians appear to overestimate the duration of life of
end of life ill cancer patients, particularly those patients who die early in the
terminal phase and who may potentially benefit from earlier participation in
palliative care programs. CES should be considered one of many criteria, rather
than a unique criterion, by which to choose therapeutic intervention or health care
programs for patients in the end of life cancer phase. Cancer 1999;86:170 – 6.
© 1999 American Cancer Society.
KEYWORDS: prognosis, survival, end of life cancer, life expectancy.
Presented in part at the 12th International Congress on Care of the Terminally Ill, Montreal, Quebec, Canada, September 13–17, 1998.
The authors are grateful to Mrs. Brenda Topliss for
her invaluable technical assistance.
Address for reprints: Maria E. Suarez-Almazor,
Ph.D., 214 Heritage Medical Research Centre, Faculty of Medicine, University of Alberta, Edmonton,
Alberta T6G 2S2, Canada.
Received September 14, 1998; revision received
January 25, 1999; accepted February 16, 1999.
© 1999 American Cancer Society
espite major medical advances, the rate of incidence and the
number of cancer deaths have increased substantially in recent
years. The number of people diagnosed with cancer in Canada has
increased by 30% in the last decade, and it is estimated that in
developed countries approximately 30% of the population alive today
will develop cancer in their lifetimes.1 For approximately 50% of
patients diagnosed with cancer, active treatments (aimed primarily at
a cure or prolongation of life) become ineffective at a certain point.2
Although the definition of life expectancy for these patients often
relies on the clinical estimation of survival (CES), it is still unclear
whether this information should be used in the counseling of patients
and families and/or in establishing patients’ eligibility for tumordirected therapies or palliative care programs. In the U.S., a 1993
report from the National Hospice Organization showed that over 50%
of patients with end of life cancer were not given access to hospice
services3 or were referred too late in the course of their illness to take
full advantage of the support provided by hospice programs.4 Overly
Survival Prediction and End of Life Cancer/Viganò et al.
optimistic CES made by different health care providers has been suggested to be among the major determinants of these late referrals.5 On the other hand,
some studies have shown the accuracy of clinical predictions of survival to be similar6 or superior7 to predictions based on the assessment of characteristics of
major prognostic importance, such as performance
status. Heterogeneity among studies could perhaps
explain some of these differences. Some authors have
indicated that the magnitude of the interaction between CES and actual survival may vary according to
the length of follow-up or the median survival of the
sample, as has been shown for other survival predictors.8 Other studies have suggested that the difficulty
in predicting terminal survival may be that death
among cancer patients is due to many essentially unpredictable events.9 –11 To define better the role of
physicians’ prediction in the management of end of
life cancer patients, we prospectively evaluated the
accuracy of their CES in a population-based cohort of
patients who were seen at the onset of their terminal
phase. We also systematically reviewed the available
literature on this subject.
An inception cohort of sequentially accrued patients
with advanced cancer was assembled at the Cross
Cancer Institute (CCI) in Edmonton, Alberta, Canada.
The Institute is the only referral center for oncologic
treatment in northern Alberta, and its catchment area
has a population of approximately 1.5 million. During
the accrual period, 270 patients who were seen as
outpatients or admitted as inpatients were potentially
eligible for the study. Patients were eligible if they
were older than 18 years and had terminal cancer of
the lung, breast, gastrointestinal system, or prostate.
Their diseases were defined as terminal when further
attempts to arrest or control progression were deemed
unavailable.12 Patients were considered eligible for the
study if they entered the terminal phase within 30 days
of their possible accrual. Of the 270 patients who met
the inclusion criteria, 248 (92%) agreed to participate
in the study. Patients were given an initial assessment
followed by monthly telephone interviews throughout
the course of their disease until death or the end of the
study. At the time of the initial assessment, the treating oncologist was asked to provide a CES in weeks or
months. The CES was recorded for 233 patients.
Statistical Analysis
The accuracy of CES was compared with the actual
survival of patients using the following methods:
1. The difference between CES and actual survival
was calculated for each patient. One-way analysis
of variance (ANOVA) was performed to compare
the mean values of these differences among different oncologists. The Bonferroni test was applied for multiple comparison procedures.13 To
determine whether prediction errors were associated with particular physicians or with patient or
disease characteristics of clinical and prognostic
value, logistic regression analyses were performed.
The dependent variable was dichotomous: accurate (within 1 month of error) versus inaccurate
(error .1 month). Independent variables were
age, gender, type of disease, type of treatments
received during and after study enrollment, performance status, and source of referral.
2. The relation between actual survival and CES was
assessed with the Pearson correlation coefficient
(r) and the intraclass correlation coefficient (ICC).
The ICC is a reliability coefficient calculated from
an ANOVA. The ICC reflects the degrees of both
correspondence and agreement between ratings,
adjusting for the role of chance in the agreement.
3. Both CES and actual survival were categorized
into 1) #2 months, 2) 2– 6 months, and 3) .6
months. These categorizations are commonly
used both in Canada and in the U.S. to provide
access to government-funded hospices and regional and hospital-based palliative care units.
The median length of stay in hospices in the U.S.
is 29 days; consequently, half of patients are referred to hospices with less than 1 month to live.14
The sensitivity, specificity, positive (PPV) and negative predictive values (NPV), and accuracy of CES
were estimated as follows: Sensitivity was calculated as the percentage of patients in a specific
survival category who were correctly predicted—
that is, the number correctly predicted within that
category divided by the total number of patients
whose actual survival was within that category.
Specificity was calculated as the number of patients correctly classified as not belonging to a
specific survival category divided by the number
of patients wrongly classified as belonging to a
specific survival category plus the number of patients correctly classified as not belonging to a
specific survival category. The PPV was calculated
as the percentage of patients, of those predicted
by the physician to be within a specific survival
category, who were correctly predicted—that is,
the number correctly predicted within that category divided by the total number of patients
whose survival was predicted to be in that category. The NPV was calculated as the percentage of
CANCER July 1, 1999 / Volume 86 / Number 1
patients who were correctly predicted not to belong to a specific survival category—that is, the
number correctly predicted not to be in that survival category divided by the total number of patients whose survival was predicted to be in that
category. Accuracy represents the percentage of
patients correctly classified by physicians within
the categories. These cutoffs were also used to
calculate an ordinal measure of agreement
(weighted kappa, k).15 Kappa is the ratio of the
number of observed nonchance agreements to the
number of possible nonchance agreements. Incremental weights were assigned to disagreement
cells in the agreement matrix, according to the
formula w 5(r1–r2)2, where w is the assigned
weight and r1–r2 represents the deviation from
agreement between CES and actual survival for
each cell in the agreement matrix. For patients
who were predicted to live between 2 and 6
months but actually lived either #2 months or .6
months, the observed and expected frequencies
were multiplied by a w of 1, whereas for patients
who were predicted to live either #2 months or .6
months but actually lived either .6 months or #2
months, respectively, w was calculated to be 4.15
Statistical significance was set at P # 0.05 (twotailed). All analysis were carried out using the SPSS 6.1
statistical software package.16
Publications included in the review of the literature on CES for advanced or end-stage cancer patients were initially identified through MEDLINE,
CANCERLIT, and EMBASE, using two search strategies (which will be provided by the authors on request) developed by a research librarian. The years
1966 –1998 for MEDLINE, the years 1983–1998 for
CANCERLIT, and the years 1980 –1998 for EMBASE
were respectively searched. We also examined the
reference lists from selected articles. Data reviewed
and extracted from selected articles included sample size, type of primary tumor, length of survival,
type of study, whether the cohort was selected at the
inception of the terminal phase, correlation coefficients (r) between actual survival and CES, and
other relevant outcomes of the studies. The above
design features are thought to be generally relevant
to the quality of studies on prognosis; the correlation coefficient was one of few elements to allow
quantitative comparisons among the studies.17–19
Table 1 shows selected demographic and clinical characteristics of the 248 patients originally included in the
study. At the time of analysis (April 1998), 225 patients
Selected Demographic and Clinical Characteristics of 248 Patients
Median age, yrs (range)
Lost to follow-up
Primary site
Tumor-directed treatmentsa
No treatment
62 (29–92)
Tumor-directed treatment was administered concurrently or after study accrual of patients, with
palliative intent.
(91.0%) had died, 23 (9.0%) were alive, and no patient
was lost to follow-up. The estimated median survival
time for the group overall was 15.3 weeks.
The difference between CES and actual survival
was calculated for the 210 patients who were known
to be dead and who had had their survival predicted. Fifteen patients had had their date of death
recorded but had not had their survival predicted.
The accuracy of CES is presented in Figure 1, which
shows the differences between CES and actual survival. The median difference between CES and actual survival was 1.1 month (range, 612 months) in
the optimistic direction. Fifty-two patients (25%)
had had their survival correctly predicted by their
treating oncologist to within 1 month, 49 (23%) had
had their survival underestimated, and 109 (52%)
had had their life expectancy overestimated. The
Pearson correlation coefficient between actual survival and CES was 0.47 (P , 0.0001) and the intraclass correlation coefficients were 0.46 and 0.47,
respectively, for random and fixed models.
No differences were noted in the accuracy of prediction by different oncologists; therefore, there was
no need to stratify the analysis further according to
physician. This was further supported by the result of
a logistic regression analysis that failed to demonstrate
any significant associations between inaccurate prediction and source of referral (medical oncologists vs.
Survival Prediction and End of Life Cancer/Viganò et al.
FIGURE 1. Difference between the
clinical estimation of survival and the
actual survival.
Overall Comparison of Clinical Estimation of Survival (CES) and Actual Survival
Actual Survival
Clinical estimation of survivala
<2 mos
2–6 mos
>6 mos
#2 mos
2–6 mos
.6 mos
23 (31%)
42 (57%)
9 (12%)
5 (6%)
53 (68%)
20 (26%)
3 (4%)
34 (42%)
44 (54%)
Clinical estimation of survivala
Positive predictive
value (PPV)
Negative predictive
value (NPV)
#2 mos
2–6 mos
.6 mos
Accuracy 5 52%; weighted kappa 5 0.42.
radiotherapy oncologists), age and gender of patients
(male vs. female, #65 vs. .65 years), type of disease
(lung vs. nonlung), treatment choice for antitumor
therapy in the terminal phase (no treatment or treatment discontinued vs. treatment initiated or continued), and performance status (bedridden vs. ambulatory).
The categorization of both actual survival and CES
allowed the inclusion of patients who were still alive at
the date of study closure. Two hundred thirty-three
patients were included in one of the above categories
(Table 2). Almost half of the patients were not assigned
to the correct survival categories, and the weighted k
only showed moderate agreement between CES and
actual survival (k 5 0.42). In particular, low sensitivity
of CES was observed for patients who died within 2
months, suggesting that physicians had some diffi-
culty identifying patients with short survival. Sensitivity was also low (54%) for patients who survived longer
than 6 months. On the contrary, specificity was higher
for shorter and longer survival. These findings show
that oncologists were more likely to predict survival
periods of between 2 and 6 months, often inaccurately. Positive predictive values in this context reflect
the probability that a given CES is accurate; they were
74% for #2 months, 41% for 2– 6 months, and 60% for
.6 months. When physicians predicted a short survival period, they were generally correct (PPV 74%),
but they only predicted this for a small proportion of
those patients who actually survived ,2 months (sensitivity 31%, overestimation 79%).
Several hundred publications were identified
searches. Of these, 11 studies were related to CES and
CANCER July 1, 1999 / Volume 86 / Number 1
Methodologic Features and Relevant Outcomes of the 11 Studies Included in the Review
Median survival
Pearson r
between As and
Clinical trial
(range) 0.24–0.28
Evans, 19856
Home care
0.44 (initial CES)
0.58 (final CES)
Heyse-Moore, 198722
Forster, 198823
(range) 0.30–0.40
Bruera, 199225
Palliative care
Maltoni, 19947
Home care
Maltoni, 199526
Home care
Muers, 199627
Oxenham, 199828
(range) 0.31–0.72
(initial CES)
(range) 0.61–
(final CES)
Study (by first
No. of
Parkes, 197220
Scotto, 197221
Other relevant outcomes
87% of errors were optimistic.
Physician could clearly identify patients with
prognoses ,2 mos or .6 mos. Equal percentage
of over estimation and underestimation.
Predictions tended to be overoptimistic. Karnofsky
performance index gave a closer correlation with
AS than CES. Final CES better than initial ones.
Most predictions were over-estimated.
High variation in performances of different health
professionals. Poor correlation coefficients among
the estimates made by different health
professionals. Serious errors tended to
overestimate survival. CES were fairly accurate
within a 3-mo boundary.
Doctors and nurses were as likely to over estimate as
to underestimate life expectancies .1 or ,1 yr.
PPV 5 0.86, NPV 5 0.64. Accuracy 5 0.68 for CES of
patients who died/survived within/beyond 4
Difference in AS-CES 5 1 wk. For 62% of patients
survival was overestimated and for 37%
underestimated. CES performed better than KPS
assessment (r 5 0.37). Better correlation AS-CES
according to level of physician experience.
CES remained statistically significant in multivariate
For 36% of patients survival time was
underestimated, for 55% overestimated.
Sensitivity 5 52%, 73%, 38%; PPV 5 80%, 50%,
49%; accuracy 5 57%, for predictions of #3, 3–9
and .9 mos. CES could differentiate further
prognostic groups in a multivariate model.
Predictions made within 1 week of death more
accurate than initial CES. Auxiliary nursing
provided better prognosticators than senior team
members when patients were near death.
AS: actual survival; CES: clinical estimation of survival; N: sample size; MS: median survival; PCU: palliative care unit; KPS: Karnofsky performance status assessment; NR: not reported; r: correlation coefficient; PPV:
positive predictive value; NPV: negative predictive value.
Eighty-three percent of the sample died within 12 weeks.
Mean survival.
were reviewed6,7,20 –28 (Table 3). These studies reported median survival periods ranging from 2.4 to 22
weeks. Only two studies examined nonsmall cell lung
carcinoma as specific primary tumor,21,27 whereas the
remaining considered any end stage malignancies.
Cohorts were not identified at the inception of the
terminal phase in any of the studies, and only 224,27 of
11 included population-based samples. In 6 of 11
studies in which this information was available, the
correlation coefficients between actual survival and
CES were in the poor-to-moderate range. Seven studies clearly reported a tendency to overestimate survival, whereas in another two,21,24 health professionals
appeared as likely to overestimate as to underestimate
life expectancies. Similarly to our findings, in which
health professionals actually predicted a short survival
period, they were fairly accurate, with moderate-tohigh PPV.25,27 Sensitivity, however, was low, because
they only predicted a short survival period in a proportion of those who actually died within the first 2–3
Survival Prediction and End of Life Cancer/Viganò et al.
months. Poor correlation among the estimates of different health professionals was reported in one
study.23 In another, CES appeared to be independent
of other survival indicators in multivariate analyses.26,27 These findings suggest that CES does not rely
on “common observations” and that it is highly dependent on the individual perceptions of the person
making the determination. The level of personal
knowledge of the patient appears to improve the prediction.6,28 It is still unclear whether CES is equally
improved by the seniority level of health professionals,
although seniority did not appear to have an impact in
our study.
The objective of this study was to evaluate the accuracy of CES in patients with end of life cancer.
Although some other studies have also assessed
clinical predictions of survival, they have generally
been based on patients admitted to hospices or
receiving palliative care in selected settings. We included in our study a population-based cohort of
patients consecutively accrued at a specific inception point, defined as the beginning of the terminal
phase as determined by the treating oncologist. As
such, our sample is more representative of patients
with end of life cancer in the population at large
than are some previous studies.
A moderate correlation was observed between actual survival and CES, with values of approximately
0.50 for the bivariate correlation coefficient, the ICC,
and the weighted k. These last two coefficients take
into account agreement by chance alone and are more
appropriate than the simple correlation coefficients
used in previous studies.18 Only 25% of the physicians’
predictions of survival were accurate within 1 month.
Two-thirds of the incorrect CES overestimated survival, particularly for those patients who had shorter
survival. Overestimation of survival has also been reported by others.25–27
Survival was categorized into 3 groups: #2 months,
2– 6 months, and .6 months. These categories were
based on cutoff values applied for the provision of palliative care services in some settings. Using these broad
categories, the accuracy of the CES improved to 52%.
These values are also consistent with those reported in
other studies that classified survival into similar
groups.25,27 In general, when physicians predicted a
short survival period, they were correct (PPV 74% for CES
#2 months). However, they only predicted this short
survival period for one-third of those who actually died
within 2 months of entering the terminal phase. A concern regarding overestimation of survival is that some
patients may be denied prompt access to palliative care
if there is an erroneous expectation of longer survival,
because some health care programs are only accessible
to patients who are expected to live for less than 2
months. In our study, physicians were also more likely to
be accurate when predicting survival periods longer than
6 months. Nevertheless, for 40% of the patients who
actually lived longer than 6 months, the treating physician had predicted a shorter survival period. Most often,
physicians expected the survival of terminally ill patients
to be within 2 and 6 months, when, in fact, one-third
survived 2 months or less, another third 2–6, months
and the remaining one-third longer than 6 months.
These findings suggest that CES on its own is not an
accurate parameter for a quantitative prognosis. However, a few studies that examined the potential value of
CES when included in prognostic models based on patient characteristics have found that CES remains an
independent survival predictor, adding some information not otherwise collected or measured.26,27
It has been suggested that CES relies on a “subconscious appraisal.”29 Our findings and review of the
literature confirm that the performance of CES is independent of the level of seniority among health professionals24,28 and of other patient and disease characteristics that have been found to be of significance
to the prognoses of terminally ill cancer patients.26,27
The major problem with CES appears to be its lack of
sensitivity in identifying patients with poorer survival
for whom an adequate delivery of palliative care
would be the most appropriate. An extensive review of
patients’ medical charts, as performed by Forster et
al.,23 or the “subconscious appraisal” that seems to
have characterized CES in our study and other studies,26,27,28 do not work well by themselves. It is possible
that a better knowledge of patients, in combination
with some other possible predictors of survival, may
improve prediction, as has been suggested in some of
the reviewed publications.6,25,28
This study and our review of the literature show
that health professionals are often inaccurate in their
survival predictions for patients with end of life cancer, and that they tend to overestimate life expectancy,
particularly for patients who die soon after entering
the terminal phase. These patients could perhaps benefit from being involved in palliative care programs at
an earlier stage. Better estimation of survival may be
achieved by adding some clinical indicators of prognostic value (i.e., performance status, nutritional parameters, etc.) to a prognostic model. On its own, CES
appears to be inadequate for making decisions in relation to the care of the terminally ill. It should be
considered one of many criteria, rather than the sole
benchmark, by which to choose a therapeutic inter-
CANCER July 1, 1999 / Volume 86 / Number 1
vention or health care program for cancer patients in
the terminal phase.
Fraumeni JF, Devesa SS, Hoover R, Kinlen L. Epidemiology
of cancer. In: De Vita V, Hellman S, Rosemberg S, editors.
Cancer principles and practice of oncology. 4th edition.
Philadelphia: J. B. Lippincott, 1994:150 – 82.
WHO Expert Committee. Cancer pain relief and palliative
care. Geneva: World Health Organization, 1990.
National Hospice Organization (NHO) Newsline. 1992 stats
show continued growth in programs and patients. NHO
Newsline 1993;3:1–2.
Demer C, Johnston-Anderson AV, Tobin R, Cohen MH, et al.
Cost of hospice care: late versus early entry [abstract]. Proc
Am Soc Clin Oncol 1992;11:392.
Kinzbrunner BM. Ethical dilemmas in hospice and palliative
care. Support Care Cancer 1995;3:28 –36.
Evans C, McCarthy M. Prognostic uncertainty in terminal
care: can the Karnofsky index help? Lancet 1985,25,1204 – 6.
Maltoni M, Nanni O, Derni S, Innocenti MP, Fabbri L, Riva
N, et al. Clinical prediction of survival is more accurate than
the Karnofsky performance status in estimating life span of
the terminally ill cancer patients. Eur J Cancer 1994;30A:6;
764 – 6.
Byar DP, Mantel N, Hankey B. Predicting survival in terminal cancer. Br Med J 1973;1:611.
Den Daas N. Estimating length of survival in end-stage
cancer: a review of the literature. J Pain Symptom Manage
1995;10:548 –55.
Klastersky J, Daneau D, Verhest A. Causes of death in patients with cancer. Eur J Cancer 1972;8:149 –54.
Inagaki J, Rodriguez V, Bodey G. Causes of death in cancer
patients. Cancer 1973;2:568 –73.
Saunders CM. Appropriate treatment, appropriate death. In:
Saunders C, editor. The management of terminal disease.
London: Arnold, 1978.
Portney LG, Watkins MP. Analysis of variance. In: Portney
LG, Watkins MP. Foundations of clinical research: applications to practice. Norwalk: Appleton & Lange, 1993:373–95.
Christakis NA. Timing of referral of terminally ill patients to
an outpatient hospice. J Gen Intern Med 1994;9:314 –9.
Portney LG, Watkins MP. Statistical measures of reliability.
In: Portney LG, Watkins MP. Foundations of clinical research: applications to practice. Norwalk: Appleton & Lange,
SPSS Inc. SPSS for Windows. Base system user’s guide. Release 6.0. Chicago: SPSS Inc.,1993:350.
Sackett DL, Haynes RB, Guyatt GH, Tugwell P. Making a
prognosis. In: Sackett DL, Haynes RB, Guyatt GH, Tugwell P,
editor. Clinical epidemiology: a basic science for clinical
medicine. 2nd edition. Boston: Little, Brown, 1991.
Laupacis A, Wells G, Richardson WS, Tugwell P, for the
Evidence-Based Medicine Working Group. Users’ guides to
the medical literature. V. How to use an article about prognosis. JAMA 1994;272:234 –7.
Sackett DL, Richardson WS, Rosemberg W, Haynes RB. Evidence-based medicine. New York: Churchill Livingstone,
Parkes CM. Accuracy of predictions of survival in later stages
of cancer. Br Med J 1972;2:29 –31.
Scotto J, Schneiderman MA. Predicting survival in terminal
cancer. Br Med J 1972;4:50.
Heyse-Moore LH, Johnson-Bell VE. Can doctors accurately
predict the life expectancy of patients with terminal cancer?
Palliat Med 1987;1:165– 6.
Forster LE, Lynn J. Predicting life span for applicants to
inpatient hospice. Arch Intern Med 1988:148;2540 –3.
Addington-Hall JM, MacDonald LD, Anderson HR. Can the
Spitzer Quality of Life Index help to reduce prognostic uncertainty in terminal care? Br J Cancer 1990,62:695–9.
Bruera E, Miller MJ, Kuehn N, MacEachern T, Hanson J.
Estimated survival of patients admitted to a palliative care
unit: a prospective study. J Pain Symptom Manage 1992;7:
82– 6.
Maltoni M, Pirovano M, Scarpi E, Marinari M, Indelli M,
Arnoldi E, et al. Prediction of survival of patients terminally
ill with cancer. Cancer 1995;75:2613–22.
Muers MF, Shelvin P, Brown J. Prognosis in lung cancer:
physicians’ opinion compared with outcome and a predictive model. Thorax 1996;51:894 –902.
Oxenham D, Cornbleet MA. Accuracy of prediction of survival by different professional groups in a hospice. Palliat
Med 1998;12:117– 8.
Holland JF. Prediction of time of death in patients with
advanced cancer. AnnN Y Acad Sci 1969;164:678 – 86.
Без категории
Размер файла
96 Кб
Пожаловаться на содержимое документа