Since the first case of acquired immune deficiency syndrome (AIDS) was detected in the United States in 1981, this disease has spread to every corner of the world. Despite the annual incidence of AIDS showing a slight decline since 1997, it remains a major cause of infectious disease-related death worldwide. Unlike global epidemiological trends, the number of notified AIDS cases has increased in recent years in China^[1]. Moreover, among the class B notifiable infectious diseases, AIDS was responsible for the highest mortality rate from 2008 to 2019, still failing to be identified as a vaccine-preventable disease^[1]. Thus, to provide more unambiguous and quantitative direction for the effective formulation of preventive planning strategies and for the reasonable allocation of limited resources, a forecasting model with high precision and accuracy to understand the future epidemic behaviors of AIDS is required. During the model-building process, accurate statistics are vital to ensure the forecasting accuracy of the model.

Since December 2019, coronavirus disease 2019 (COVID-19) has rapidly evolved into a major global public health issue. Confronted with this unprecedented crisis, countries have taken various measures to stop the continued spread of COVID-19. In China, a series of effective measures were implemented, and these measures played an important role in stopping the spread of COVID-19^[2]. However, it is discovered that these rigorous prevention and control strategies and measures also had an effect on the incidence and mortality rates of other diseases; however, further studies are needed to measure the impacts. To date, there are no reports on the effect of such interventions on AIDS case numbers in China.

Time series analysis is frequently used to provide predictions that aid the development of effective prevention and control strategies. The causal impacts of strict counter-virus measures on AIDS case numbers can be assessed with intervention analysis under the Bayesian structural time series (BSTS) model that can generate a counterfactual prediction in a synthetic control series to describe outcomes in the absence of the intervention^[3]. Therefore, the current study aims to estimate the effect of counter-COVID-19 measures on the reduction in AIDS case notifications and to predict the epidemic patterns using the BSTS model. Our findings offer insight that will aid the development of effective strategies to control AIDS case numbers in China in the future.

The monthly number of AIDS cases and population data from 31 provinces in mainland China from January 2004 to December 2020 were collected. All cases were checked medically and diagnosed in accordance with the diagnostic criteria for HIV/AIDS before being reported to the Chinese Centre for Disease Control. Considering the effects of COVID-19 on the epidemiological trend of AIDS, the data from January 2004 to December 2019 were only used when exploring the usefulness and flexibility of the BSTS model in forecasting trends in the AIDS epidemics. The average annual percentage change (AAPC) and annual percentage change (APC) were employed to describe the epidemic patterns of AIDS^[4]. The BSTS model was used to perform causal inference and time series forecasting. The mean absolute deviation (MAD), root mean square error (RMSE), mean absolute percentage error (MAPE), mean error rate (MER), and root mean square percentage error (RMSPE) were adopted to judge the forecasting performance of the BSTS model^[4]. All statistical analyses were performed using R software (Version 3.4.3).

During the study period, there were a total of 643,800 case notifications (2.866 per 100,000 persons). A notable upward trend was observed in case numbers [AAPC = 20.446, 95% confidence interval (CI) 16.685 to 24.329; t = 11.493, P < 0.001] that comprised two stages: an upsurge from 2004 to 2010 (APC = 41.731, 95% CI 31.421 to 52.849; t = 10.062, P < 0.001) and then a slow increase from 2011 to 2020 (APC = 6.127, 95% CI 3.825 to 8.481; t = 5.907, P < 0.001) (Supplementary Figure S1, available in www.besjournal.com). The primary reason for this variation may be China's action planning on the prevention and control of AIDS (2006–2010), which stated that by 2010, the number of people living with HIV in China must be reduced to less than 1.5 million, and therefore a series of AIDS prevention measures were implemented. The incidence of AIDS showed a trough in January/February and a peak in December each year, and this seasonal pattern became more marked since 2010 (Supplementary Figure S2, available in www.besjournal.com). The effect of the spring festival may be closely related to the seasonality of AIDS.

Figure S1.  Joinpoint regression plot showing the AIDS epidemiological trends during 2004–2020. ^*Showed that the annual percent change (APC) is statistically significant.

Figure S2.  Time series plot showing the AIDS incidence and the decomposed tend and cyclicity based on the Hodrick-Prescott filter technique.

The year 2020 was regarded as a period of intervention when performing intervention analysis using the BSTS method. In estimating the causal effect of interventions on case number reduction by use of the BSTS method, the seasonality, time effects, and the population were deemed as covariates. The predicted expected case notifications are provided in Table 1, suggesting a monthly average decrease in AIDS case notifications of 41% (95% CI 24% to 58%) from January–March 2020, 19% (95% CI 14% to 46%) from January–June 2020, and 16% (95% CI 7.3% to 24%) from January–December 2020 owing to the COVID-19 pandemic. The BSTS model can infer the temporal changing trends of potential effects, incorporate prior empirical information on the parameters, and deal with a complex covariate structure^[3]. Considering that the Bayesian confidential interval and posterior probability of a causal effect under the BSTS model rejected the reduction in AIDS case notifications owing to the COVID-19 outbreak as a random event, this confirmed reliable evidence for a true causal impact (Table 1, Supplementary Figures S3 and S4, available in www.besjournal.com). This result suggested that COVID-19 may have resulted in medium- and longer-term impacts for the AIDS epidemic, which corroborated a previous finding that COVID-19 may have medium- and longer-term impacts on disease patterns^[5]. There are a number of possible explanations for these effects^{[5, 6]}. First, people with chronic conditions or mild symptoms were discouraged from seeking medical help to reduce congestion in hospitals. Second, the stringent prevention and control measures and the mandatory requirement for the negative results of nucleic acid testing may have resulted in reluctance among people infected with HIV to have a medical diagnosis and examination. Third, many hospitals declined non-urgent service requests, especially when the medical system was overburdened during the peak of the COVID-19 outbreak. Fourth, there has been a significant decrease in the number of people administered preventive AIDS treatment, and a decrease in spending on AIDS diagnostic, therapeutic, and prevention services. Fifth, AIDS-specialized medical personnel and molecular diagnostic platforms may have been reallocated to the COVID-19 response. Sixth, the stringent anti-contagion policies restricted the movement of people, and the sudden economic disruption may have made it difficult for people infected with HIV to visit hospitals. Seventh, there are some similarities between the clinical features of AIDS (e.g., influenza-like symptoms, systemic fatigue, loss of appetite, fever, and cough) and COVID-19, resulting in concerns about stigma. Lastly, there have been delays in reporting HIV infections during the COVID-19 pandemic.

The expected figures were forecasted by averaging across the 500 Markov Chain Monte Carlo (MCMC) draws under the BSTS model (Supplementary Tables S1–S6, available in www.besjournal.com). Table 2 lists the values of MAD, RMSE, MAPE, MER, and RMSPE, which measure the predictive ability of the BSTS model. Accurate forecasts for the 12, 24, 36, 48, and 60 holdout periods were obtained because of the low error rates (this can be attributed to the BSTS model accurately showing the stochastic behavior of a time series and generating a projection based on the Bayesian model averages^[7]). Whereas, an unacceptable forecast for the 72 holdout period was obtained owing to the large values of the above-mentioned five indices, especially the MAPE value, which was greater than 50% (Supplementary Figure S5, available in www.besjournal.com).

At present, the WHO has proposed an ambitious target for AIDS prevention, with a 75% reduction in new infections, including key populations, from 2010 to 2020, and the end of the AIDS epidemic as a public health threat by 2030^[8]. Accordingly, we re-erected the BSTS model based on 17 complete years of data (including the modified 2020 data) to forecast the AIDS epidemic from January 2021 to December 2025. The results showed that the estimated incident rates will continue to surge (APC = 6.374, 95% CI 5.541 to 7.213; t = 16.855, P < 0.001) reaching a comparatively high level of 94,870 cases (95% CI 1,620 to 5,611,911) in 2025, which will be 2.593-fold higher than the 36,594 cases recorded in 2010 (Supplementary Table S7, Figures S6 and S7, available in www.besjournal.com). Currently, an estimated 70% of individuals infected with HIV globally occur in low- and middle-income settings, mainly in eastern and southern Africa, western and central Africa, and the Asian–Pacific region^[9], and achieving set objectives will also depend on whether efforts in these settings accelerate or stall. Moreover, China's new AIDS cases accounted for 23% of the total number of new infections in the Asian–Pacific region in 2019^[9]. These findings indicate that AIDS will still plague China and achieving the goal of ending the AIDS epidemic by 2030 poses a great challenge^[8]. Therefore, to tackle such a major public health problem, intervention strategies need to be strengthened and new effective prophylaxis methods for AIDS need to be proactively explored. In addition, it is estimated that more than 60% of AIDS cases are infected through heterosexual sex, and student infections are increasing in China^[10], thus there is an imminent need to implement HIV screening and popularize earlier sex education (from age 9) according to UNESCO guidance.

Figure S6.  Time series plot showing the forecasting results until 2025. This plot displays the actual series in red, and the fitting and forecasting results as a blue line. It seemed that the BSTS method predicted a rising trend from January 2021 to December 2025.

Figure S7.  Joinpoint regression plot showing the AIDS epidemiological trends during 2004–2025. ^*Showed that the annual percent change (APC) is statistically significant.

This study had several limitations. First, the data spanned the period of policy intervention (2010). Though the results stemmed from the preferred model are as good as the expected, the real effects before and after the sudden escalation of AIDS morbidity on the model accuracy remain unknown. Second, an estimated 32.4% of people who have been infected with AIDS are unaware of their infection status in China^[1]. Hence, the actual situation may be more serious than that estimated. Third, many complicated influencing factors, some of which may be unpredictable, can be contributors to the transmission of AIDS incidence. Fourth, it is necessary to make timely updates to models as short-term data becomes available to increase the predictive ability of the model. Lastly, the analytical epidemic data merely represent the country-level situation and overall trends in AIDS incidence, our method may be of great benefit to other regions or infectious diseases as well.

In summary, the current findings indicate that COVID-19 may have significant consequences for the AIDS epidemic, as determined by the BSTS model. This model can act as a useful technique to estimate the temporal behaviors of AIDS, and thus assist policymakers in rationally allocating health resources and appropriately formulating prevention and control strategies for this disease. Since new AIDS infections will continue to increase, there is an urgency to adopt appropriate and effective prevention and control policies in China.

Figure S3.  Time series plot displaying the causal impacts of the COVID-19 outbreak on the decreases in AIDS cases from January–December 2020. The first panel provides the reported AIDS cases and counterfactual expected figures for the post-outbreak period. The second panel describes the pointwise causal impact that indicates the difference between reported cases and expected figures. The third panel illustrates a cumulative effect of the COVID-19 pandemic on the decreases in AIDS cases via accumulating the pointwise effects from the second panel.

Figure S4.  Contribution of trend, seasonality, and regression (COVID-19 outbreak) components to the AIDS incidence. As depicted, the AIDS incidence has a noticeable seasonal pattern and the COVID-19 outbreak causes a remarkable decrease in AIDS cases during this period.

Figure S5.  Time series plot showing the forecasted temporal trends of the AIDS incidence for (A) 12 holdout data, (B) 24 holdout data, (C) 36 holdout data, (D) 48 holdout data, (E) 60 holdout data, and (F) 72 holdout data under BSTS approaches. It can be seen that the BSTS approaches are able to better capture the long-term epidemiological trends of AIDS for the 60-step ahead forecast except for the 72 holdout data.