Comparative Analysis and Prediction of Malignant Tumor Mortality

in China Based on LSTM Models

Fan Yang

School of Computer Science and Engineering, Central South University, Nan Chang, China

Keywords: Malignant Tumor, LSTM Model, Mortality Rate.

Abstract: The prediction and analysis of the mortality rate of malignant tumors is a hot topic of great importance and

urgency in the world. However, the dynamic mortality prediction model commonly used in the medical field

still has some limitations in the nonlinear structure of mortality research. Therefore, based on the survey data

of "Mortality rate of major diseases in selected urban/rural areas of China" from 2008 to 2021, this research

will utilize the Long and short-term memory networks (LSTM) model to validate the model and predict the

mortality rate in 2022. This research will also quantitatively analyze and explore the reasons for the mortality

rate differences arising from regional differences and gender differences. The prediction results show that the

mortality rate in urban areas in 2022 may increase compared with the value in 2021, while the mortality rate

in rural areas will decrease, and the difference in the trend is only reflected in the regional differences.

Furthermore, the trends in mortality rates over the years show a general decline in urban areas and an increase

in rural areas, with the urban mortality rate being lower than the rural rate after 2020, and the male mortality

rate being much higher than the female mortality rate.

1 INTRODUCTION

Malignant tumors, one of the most dangerous diseases

nowadays, always occupy the first place on the list of

causes of death in many countries (Sung et al. 2021).

Rapid economic development, increasing

environmental pollution, and changes in people's

lifestyles have increased the incidence of malignant

tumors. China National Cancer Center released a new

report on malignant tumor statistics, which reveals

that the number of new cases of malignant tumors in

China reached 4,064,000 in 2016, and the total

number of deaths reached 2,414,000 (Rongshou et al.

2022). The development of society and the economy

are now seriously hampered by malignant tumors,

which also pose a serious threat to people's lives and

health. The importance and urgency of malignant

tumor-related research have become self-evident. The

mortality rate is an important index in malignant

tumor-related research, and the prediction and analysis

of malignant tumor mortality rate is also a hot topic.

Commonly used mortality prediction models are static

and dynamic mortality prediction models. Currently,

the dynamic mortality prediction model represented by

the Lee-Carter model is commonly used in the medical

field. However, the dynamic mortality prediction

model can only portray the linear relationship between

the influencing factors and the mortality rate, and there

are some limitations in the nonlinear structure of the

mortality rate research (Gang et al. 2022). When it

comes to machine learning algorithms, the nonlinearity

of the kernel function and activation function can

realize the nonlinear mapping of the data, which

enables machine learning algorithms to have excellent

nonlinear learning ability. Moreover, for the time series

data, the Long and short-term memory networks

(LSTM) model always performs excellently. In this

research, the prediction of malignant tumor mortality

and the analysis of regional and gender differences will

help understand and predict the development trend of

malignant tumors in China, explore the sore points of

malignant tumor deaths in China, and provide targeted

preventive suggestions to the public from a multi-level

perspective.

Based on the data from the "Mortality rate of major

diseases in selected urban/rural areas of China" from

2008 to 2021, this research will validate the model and

predict the mortality rate of malignant tumors in

selected urban/rural areas of China in 2022 with the

LSTM model. The paper will also analyze the

178

Yang, F.

Comparative Analysis and Prediction of Malignant Tumor Mortality in China Based on LSTM Models.

DOI: 10.5220/0012816200003885

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Data Analysis and Machine Learning (DAML 2023), pages 178-183

ISBN: 978-989-758-705-4

differences in mortality rate by region and gender and

the reasons for these differences. This study might be

helpful for the Chinese public and medical institutions

to understand and treat malignant tumors. This

research will first use the data of the previous 12 years

to predict the mortality rate in 2020 and 2021, thus

verifying the LSTM model's applicability. Then, the

LSTM model will be used to predict the total mortality

rate in urban and rural areas and the mortality rate of

each gender in 2022. Furthermore, the differences in

mortality rate due to regional and gender differences

will be quantitatively analyzed and the reasons for

these differences will be investigated.

2 METHOD

2.1 Data Source

The data for this research come from a survey

"Mortality rate of major diseases in selected

urban/rural areas of China", published in the China

Statistical Yearbook by the National Bureau of

Statistics of the People's Republic of China (NBS).

The survey utilized a stratified probability sampling

method proportional to population size. Several large

cities and small and medium-sized cities were selected

as urban respondents, and several municipalities and

county-level cities were selected as rural respondents.

The survey conducted gender-specific mortality rate

and death cause statistics on the top 10 major diseases

among Chinese residents in recent years according to

the International Classification and Statistics

Standard-10 (ICD-10) for Diseases. Based on the

theme of this study and the distribution of causes of

death, this study excluded diseases that were not

continuously listed as the top 10 major diseases in the

survey, as well as diseases with small mortality values

ranging from 0.01% to 1%. This study will use the

mortality rate value in selected urban/rural areas from

2008 to 2021 as research data. Since malignant tumors

account for the highest proportion of causes of death

among the ten major diseases throughout the year, and

the mortality rate is also relatively high, it is of great

research significance and representativeness.

The data of this study include six indicators,

including the total mortality rate of malignant tumors,

male mortality rate, and female mortality rate in urban

and rural areas, etc.

2.2 Model

Long and short-term memory networks are a specific

kind of recurrent neural network (RNN). RNN is the

most natural architecture for sequential data due to the

chaining feature. However, since the same function

will be combined with itself many times, RNN will

face the problem of gradient vanishing or gradient

explosion and long-term dependency during training.

The long-term dependence issue brought on by RNN

will be partially resolved by Hochreiter’s unique unit

design of the LSTM model, which also effectively

realizes the modeling of time-series data (Hochreiter

and Schmidhuber 1997). RNN models all have some

repeating neural network modules, the traditional RNN

only simply repeats a single neural network layer,

while the LSTM repeats module comprises four

interacting layers, including three sigmoid layers and

one tanh layer, whose interactions tend to be very

specific.

Figure 1: Cell of LSTM model.

As shown in Fig. 1, each LSTM cell contains three

gate structures, the forgetting gate(f

), the input gate(i

and the output gate(o

), as well as a cell state to

maintain and update the state (Han et al 2021). Where

the forget gate is used to choose which information

from the cell state to discard, the structure reads the

previous output h

t-1

and the current input x

, then

conducts a Sigmoid nonlinear mapping and outputs a

vector f

, which can be described as:

( )

1t fh t fx t f

f W h W x b

−

= + +

 ()

where W denotes the weights and b denotes the bias.

What information will remain in the cell state is

determined by the input gate. It is divided into two

sections. The sigmoid layer will first determine which

values to update; the tanh layer will then generate a

new vector of candidate values C

and add it to the cell

state. The relationship is stated as follows:

( )

1t ih t ix t i

i W h W x b

−

= + +

 ()

( )

tanh

t ch t cx t c

C W h W x b

−

= + +

 ()

Next, the cell state will be updated from C

t-1

to C

as an input to the next cell, and the relationship can be

expressed as:

1t t t t t

C f C i C

−

 ()

Comparative Analysis and Prediction of Malignant Tumor Mortality in China Based on LSTM Models

179

where



denotes the Hadamard product.

The output gate is in charge of calculating the

output value depending on the cell state. To begin, use

a sigmoid layer to determine which part of the cell

state will be output. Simultaneously, the cell state is

processed by the tanh layer and multiplied by the

output of the sigmoid gate, generating the value h

which is important for the computation of the

following cell. This relationship can be stated as

follows:

( )

1t oh t ox t o

o W h W x b



−

= + +

 ()

( )

tanh

t t t

h o C=

 ()

3 RESULT

3.1 Verification

To validate the LSTM model's applicability for

forecasting malignant tumor mortality, this research

will use the malignant tumor mortality data of urban

areas from 2008 to 2019 to predict the corresponding

mortality rates in 2020 and 2021. The predictions will

also be compared with the true values to evaluate and

validate the model's applicability based on Mean

Squared Error (MSE). The models for applicability

validation will include five mainstream regression

prediction models: random forest, linear regression,

decision tree, support vector machines (SVM), and

LSTM. The prediction results are shown in Table 1:

Table 1: Mse of the Models.

Mode

Rando

Forest

Linear

Regressio

Decisio

Tree

SVM

LST

MSE

7.012

9.179

1.542

8.32

0.813

The MSE obtained by the LSTM model is 0.813,

whose magnitude is much smaller than the range of

prediction values from 150 to 180, indicating good

accuracy. Meanwhile, the MSE value of the LSTM

model is much smaller than the corresponding values

of the other four mainstream regression prediction

models, which also indicates that the LSTM model

has a more excellent performance than the other

models in this research topic. Thus, it can be verified

that the LSTM model is appropriate for the following

prediction of malignant tumor mortality.

3.2 Prediction

3.2.1 Prediction of Urban and Rural Total

Mortality Rates in 2022

In this research, the mortality rate of malignant

tumors in urban and rural areas in 2022 will be

predicted using the LSTM model. The predicted

value of the total malignant tumor mortality rate in

the urban region is 163.076, which is an increase of

2.76% compared to 2021. The predicted value of the

total mortality rate for rural regions in 2022 is

154.456, which is a decrease of 7.54% compared to

2021. At the same time, urban regions' forecasts are

5.5% higher than rural regions' forecasts. The trend of

the total mortality rate of malignant tumors in urban

and rural regions from 2008 to 2022 is shown in Fig.

1, where the horizontal coordinate indicates the year

and the vertical coordinate is the mortality rate

(1/100,000):

Figure 2: Overall urban and rural mortality rates (Picture

credit: Original).

Urban areas show a general downward trend in the

total mortality rate, with a predicted decrease of 2.33%

in 2022 compared to 2008. Rural areas show a more

mixed picture, with a 1.45% decrease in 2022

compared to 2008. Comparing the mortality rate of

urban areas with the rural ones, the mortality rate of

urban areas was higher than the rural areas until 2020,

after which it reversed. Furthermore, between 2021

and 2022, urban and rural regions are expected to

follow opposing trends, with urban areas rising and

rural areas falling.

3.2.2 Predictions of Male/Female Mortality

Rates in Urban and Rural Areas in

2022

Urban Area: The predicted male mortality rate for the

urban area is 204.753, an increase of 2.33% from

2021. While the female mortality rate for the urban

DAML 2023 - International Conference on Data Analysis and Machine Learning

180

area in 2022 is predicted to be 119.984, an increase of

2.76% from 2021. At the same time, the predicted

value of the male mortality rate is 70.65% higher than

the predicted value of the female mortality rate. The

trend of male and female mortality rates in the urban

area from 2008 to 2022 is shown in Figure 2, where

the horizontal coordinate indicates the year and the

vertical coordinate is the mortality rate (1/100,000):

Figure 3: Male/female mortality rates in urban areas

(Picture credit: Original).

Trends in male mortality rates are generally

consistent with those of female mortality rates, both

of which are on a downward trend. In the case of men,

the predicted value of the mortality rate in 2022 is

0.37% lower than in 2008. The predicted value of the

female mortality rate in 2022 is 7.15% lower than that

in 2008. Male mortality is much greater than female

mortality, and both are predicted to follow a similar

upward trend between 2021 and 2022.

Rural Area: The predicted male mortality rate in

rural areas is 197.797, which is 7.19% lower than the

value in 2021. The predicted female mortality rate in

rural areas in 2022 is 111.543, which is 6.35% lower

than that in 2021. At the same time, the predicted

value of the male mortality rate is 77.33% higher than

the predicted value of the female mortality rate. The

trend of male and female mortality rates in rural areas

from 2008 to 2022 is shown in Figure 3, where the

horizontal coordinate indicates the year and the

vertical coordinate is the mortality rate (1/100,000).

Trends in male mortality rates are generally

consistent with those of female mortality rates, both of

which are on an upward trend. In the case of males,

the predicted value of the mortality rate in 2022 is

3.32% lower than in 2008. The female mortality rate

is predicted to increase by 4.19% in 2022 compared to

2008. The male mortality rate is much higher than the

female mortality rate, and both are predicted to follow

a similar downward trend between 2021 and 2022.

Figure 4: Male/female mortality rates in rural areas (Picture

credit: Original).

4 DISCUSSION

The predicted trends in mortality rates for males and

females in the different regions are consistent with the

predicted trends of the total mortality rates in the

corresponding regions, and the differences in the

trends are only related to the differences in regional

distribution, and not so much to the differences in

gender. In general, the mortality rate in urban areas is

decreasing, probably due to the development of

medicine and increased health awareness. In rural

areas, the mortality rate is on the rise, probably

because the rural population aging problem is very

serious (Rongshou et al 2018). Until 2020, the total

mortality rate in urban regions was greater than in

rural regions, but this trend reversed after 2020. This

might be associated with the Human Development

Index (HDI), a comprehensive assessment of regional

human development in terms of health, education,

and income, with free human development as the core

concept (UNDP 1990). Research has shown that the

HDI index is closely related to malignant tumors, and

Khazaei discovered a positive correlation between the

incidence of prostate cancer and the HDI index, and a

negative correlation between the standardized death

rate and the HDI index (Khzazei et al 2016). The HDI

index in urban areas is higher than the value in rural

areas, and rural areas lack medical services and

cancer preventive understanding. All these factors

can lead to the reversal of the mortality rate in rural

areas and urban areas. The government can publicize

cancer prevention in rural areas and promote

healthcare reform to alleviate the imbalance of

healthcare resources.

As for gender differences, the mortality rate of

men is much higher than that of women, which may

be due to the following reasons: (1) males and women

have distinct physiological structures, and women

Comparative Analysis and Prediction of Malignant Tumor Mortality in China Based on LSTM Models

181

have greater estrogen levels than males, which may

be a preventive factor for the development of some

malignancies (Xiyi 2017); (2) the awareness of tumor

prevention of women is stronger than that of men (Ni

et al2019); (3) the level of exposure to risk factors of

men is higher than that of women. For example,

tobacco consumption and alcoholism are higher in

men than in women. Chronic disorders such as

hypertension and diabetes mellitus are also more

prevalent in males than in women (Ye et al 2016 &

Zengwu et l 2018). Moreover, there are differences in

the cancer spectrum of different genders, for example,

breast cancer and thyroid cancer are highly prevalent

in women but have a better prognosis, whereas

digestive tract cancers are highly prevalent in men but

have a poorer prognosis, which may lead to a higher

burden of malignant tumors in men than in women

(Wei et al 2021). Therefore, men should raise their

awareness of cancer protection and reduce their

exposure to harmful substances such as tobacco and

alcohol, which can effectively prevent cancer.

This research still has some limitations. First, the

LSTM model requires a large amount of training data

for training and prediction. Still, this research only

studied the mortality-related data of malignant tumors

during the 14-year period from 2008 to 2021, which

is a small amount of data, and it may lead to

inaccurate predictions. Second, the focus of the

analysis in this research was mainly on regional and

gender differences, and many other influencing

factors were not taken into account, such as age,

cancer spectrum distribution, and other factors. In the

future, the model and analysis can be further refined

in the above aspects and more detailed data are also

indispensable to facilitate in-depth research on this

topic.

5 CONCLUSION

The predictions in this research indicate that the total

mortality rate and the male and female mortality rates

in urban areas are likely to increase in 2022, whereas

the corresponding mortality rates in rural areas are

likely to decrease, and the differences in the trends are

only related to regional differences. Second, the

research analyzed the differences in the mortality rate

by region and gender. For regional differences, the

mortality rate in urban is declining, while the

mortality rate in rural regions is growing, and after

2020, the value of the mortality rate in urban began to

be lower than the corresponding mortality rate in rural

areas. Possible explanations for the difference include

a higher HDI index in urban regions, a deeper

understanding of tumor prevention in urban areas

than in rural ones, and a more serious population

aging problem in rural areas. The difference by

gender is reflected in the fact that the mortality rate

for men is much higher than that for women. The

reasons for this difference may include the different

physiological structures of men and women, the

higher awareness of tumor prevention in women than

in males, the higher level of exposure to risk factors

in men than in women, and the difference in the

cancer spectrum. The analysis of the malignant tumor

mortality rate, regional differences, and gender

differences in this study can help to understand and

predict the development trend of malignant tumors in

China, uncover the pain points of malignant tumor

deaths, and provide targeted preventive advice to the

public from a multi-level perspective. However, this

study also suffers from the limitations of too little data

and not enough comprehensive investigation of

factors. In the future, this research should make use

of more sufficient data and improved models to

include more factors into the investigation and

conduct a multi-factor comprehensive analysis.

REFERENCES

H. Sung, J. Ferlay, R. L. Siegel, et al., “Global cancer

statistics 2020: GLOBOCAN estimates of incidence

and mortality worldwide for 36 cancers in 185

countries,” CA: a cancer journal for clinicians, vol. 71,

pp. 209-249, 2021.

Z. Rongshou, Z. Siwei, Z. Hongmei, et al., “Cancer

incidence and mortality in China, 2016,” Journal of the

National Cancer Center, vol. 2, pp. 1-9, 2022.

Y. Gang, Y. Yanping, S. Chao, “Prediction of Mortality of

Elderly Population with an Improved AE-LSTM

Model,” Mathematical Theory and Applications, vol.

42, pp. 100, 2022.

S. Hochreiter, J. Schmidhuber, “Long short-term memory,”

Neural computation, vol. 9, pp. 1735-1780, 1997.

G. Han, S. Renju, M. Li, L. Wenyu, “A heart failure

mortality prediction model based on AB-CNN-

BILSTM,” Computer Applications and Software, vol.

38, pp. 37-42, 2021.

Z. Rongshou, G. Xiuying, L. Xueting, et al., “Analysis on

the trend of cancer incidence and age change in cancer

registry areas of China, 2000 to 2014,” Chinese journal

of preventive medicine, vol. 52, pp. 593-600, 2018.

UNDP, “Human development report 1990,” New York and

Oxford: Oxford University Press, pp. 10, 1990.

S. Khazaei, S. Rezaeian, E. Ayubi, et al., “Global prostate

cancer incidence and mortality rates according to the

human development index,” Asian Pacific Journal of

Cancer Prevention, vol. 17, pp. 3791-3794, 2016.

J. Xiyi, H. Yunqing, Y. Ding, L. Qilong, C. Kun, J.

Mingjuan, “Disparities of sex on cancer incidence and

DAML 2023 - International Conference on Data Analysis and Machine Learning

182

mortality in Jiashan county, Zhejiang province, 1990-

2014,” Chinese journal of Epidemiology, vol. 38, pp.

772-778, 2017.

W.Ni, H. KaiYong, Y. Li, “Analysis on health literacy and

its influencing factors about cancer prevention and

control among urban residents in Guangxi,” Chinese

Journal of Disease Control & Prevention, vol.23, pp.

711-716, 2019.

R. Ye, Y. Qinghua, X. Jiying, et al., “Epidemiology of

diabetes in adults aged 35 and older from Shanghai,

China.” Biomedical and Environmental Sciences, vol.

29, pp. 408-416, 2016.

W. Zengwu, C. Zuo, Z. Linfeng, et al., “Status of

hypertension in China: results from the China

hypertension survey, 2012–2015,” Circulation, vol. 137,

pp. 2344-2356, 2018.

C. Wei, C. Hongda, Y. Yiwen, L. Ni, C. Wanqing,

“Changing profiles of cancer burden worldwide and in

China: a secondary analysis of the global cancer

statistics 2020,” Chinese medical journal, vol. 134, pp.

783-791, 2021.

Comparative Analysis and Prediction of Malignant Tumor Mortality in China Based on LSTM Models

183