Topics New Drugs and the Probability of Care: Prevention of Disease Severity and Prolonged Disease

Printable PDF

Junichi Nishimura, Visiting Fellow, Pharmaceutical Industry Policy Institute, Professor, Gakushuin University
Sadao Nagaoka, Director, Pharmaceutical Industry Policy Institute, Professor, Tokyo Keizai University

New drugs create not only medical value in the treatment of diseases, but also social value, such as reducing the workload of healthcare professionals by alleviating symptoms, or enabling patients to work in good health after being discharged from the hospital due to a ^cure1). For example, when specific diseases were recalled, the social values considered important changed depending on the disease, with the importance of "reducing the burden of nursing care (reducing the physical, mental, and economic burden on family caregivers and others who provide care and support for the sick)" increasing, especially for diseases such as rheumatoid arthritis, cancer, and ^dementia2). In addition, in the Policy Research Institute News No. 65, based on the "National Survey on Basic Living Conditions," the level of caregiving (classification of care required, etc.) was used as a complementary indicator of healthy life expectancy of the elderly, and its change over time was confirmed, and by combining it with data from the Japan Health Sciences Foundation, the contribution of drugs to the level of caregiving was discussed3). ³⁾.

New drugs have the potential to decrease the burden of caregiving by preventing the severity of disease, but on the other hand, new drugs may also increase the burden of caregiving in the short term by helping critically ill patients recover. Therefore, the extent to which new drugs affect the probability of patient care and whether they lead to less prolonged hospitalization is an issue that should be clarified empirically, but there have been very few empirical analyses focusing on the probability of care for hospitalized patients due to data limitations. This paper constructs disease-specific panel data linking disease and caregiving and thereby clarifies the contribution of new drugs to the probability of caregiving. The author has analyzed the contribution of new drugs from the perspective of extending the life expectancy of the population, reducing the average length of hospital stay of hospitalized patients, and the cure rate of hospitalized patients (Policy Research Institute News Nos. 36, 37, 45, and 55) ⁴⁾, and basically the same analysis framework will be used. New drugs are expected to reduce the probability of long-term care both (1) by reducing the probability of developing diseases that may cause long-term care (e.g., reducing the probability of stroke) and (2) by reducing the probability of becoming a long-term care patient by avoiding serious and prolonged illness even if it occurs (e.g., preventing rheumatoid arthritis from becoming serious). In this paper, we focus on the effect of the new drug (2) on patients hospitalized in long-term care facilities.

Objectives and Methods of Analysis

^{Using the} Ministry of Health, Labour and Welfare's "Patient Survey" and IQVIA's International Classification of Diseases (ICD-10) drug prescription ^data5), this paper examines the following three main ^points6) for patients hospitalized in convalescent care-type wards and other facilities. First, with regard to the "physical and mental status" of hospitalized patients, we will pay particular attention to "transferring," "food intake," "swallowing," and "cleaning up after defecation," calculate the probability (%) of care for hospitalized patients in these activities by International Classification of Diseases, and analyze the relationship with the increase in drug stocks in the relevant disease field. Second, we will analyze the relationship between the average length of hospital stay of inpatients in medical care beds and the increase in the stock of medicines in the relevant disease field, using data by International Classification of Diseases in the same way as for the nursing care probability. Through this analysis, the effectiveness of new drugs in curbing the seriousness and prolongation (of symptoms and hospitalization) of patients hospitalized in convalescent care-type sickbeds, etc. will be verified. Finally, the number of inpatients admitted to care-type sickbeds is divided by the number of inpatients admitted to all sickbeds, and the ratio (%) of the number of inpatients admitted to care-type sickbeds is calculated by International Classification of Diseases to analyze the relationship with the increase in the stock of drugs in the disease field in question. This is to verify the effectiveness of new drugs in controlling the number of hospitalized patients who require long-term medical treatment. By empirically analyzing these data, we aim to provide objective evidence on the social value of pharmaceuticals, such as reducing the burden of nursing care and medical care. The following section describes the analytical model and estimation methods.

In this paper, the following three estimations are made.

In the analytical model, i is the International Classification of Diseases, t is the year (1999, 2002, 2005), and α, β, and γ for each variable represent the coefficient values (parameters) to be estimated. Although the estimation includes an error term, it is omitted in the model equation. First, in equation (1), the probability (%) of being a caregiver was calculated and used as the explained variable (outcome) by International Classification of Diseases and by cleanup of transfers, food intake, swallowing, and defecation in each year. （In formula (2), the average length of hospital stay for patients discharged from medical care beds, etc. (logarithmic value) was used. （In formula (3), the ratio (%) of the number of inpatients admitted to care-type sickbeds, etc. to the total number of inpatients was used. The explanatory variables (causes) are the same in equations (1) through (3): drug stock, importance of science, rate of surgery, and average age of hospitalized patients. These variables are also calculated by International Classification of Diseases, each year, and are explained in detail in the next section.

The coefficient values for each of the analytical models (1) through (3) are obtained by panel fixed effects estimation. International Classification of Diseases dummies are included as disease-specific effects, and year dummies are included as year-specific effects. Disease-specific effects can control for disease-specific characteristics that are difficult to ascertain directly. For example, more new drugs tend to be developed at the same time for diseases that are more likely to result in long-term care or are more severe, and including endemic effects in the model allows for a more unbiased estimation of the impact of drug stocks on the probability of care. If one does not control for endemic effects, the coefficient values are subject to estimation bias due to the missing important variables in the analytical model. Alternatively, the model may capture reverse causality, whereby the probability of care is higher for more severe diseases, which in turn encourages the development of new drugs to treat them. These estimation biases due to missing variables and reverse causality are said to be endogeneity problems, which can be mitigated by including disease-specific effects in the model. Similarly, including year-specific effects in the model allows us to account for the effects of time-varying shocks that are not observed as variables. For example, the year-specific effects will correct for estimation bias in the explanatory variables by absorbing the effects of interannual changes in the health care system, composition of hospitalized patients, etc. over the estimation period.

Each panel fixed effects estimation was weighted by the number of inpatients by international classification of diseases. This weighting makes it easier to reflect trends in the data for disease areas with large numbers of hospitalized patients. For example, it would be problematic to equate a disease area with only one hospitalized patient with a disease area with tens of thousands of hospitalized patients. The number of hospitalized patients (sample size) observed in each disease field differs, resulting in different data variability, and the stability of the estimation is increased by more easily reflecting the results of the disease field with more hospitalized patients in the data. Therefore, we attempted to estimate equations (1) and (2) weighted by the number of hospitalized patients in care-type beds, etc., and equation (3) weighted by the number of hospitalized patients in all hospital beds.

Finally, we discuss some points to be noted in the estimation. The patient survey used in this paper was conducted by stratified random sampling of medical facilities nationwide from September to October each year, and the estimated number of patients, etc., was obtained by calculating the estimated number of patients from the survey response data. Therefore, this is not a continuous survey of specific patients. Therefore, it is expected that the number of hospitalized patients includes those who have just been hospitalized at the time of the survey, those who have been hospitalized for a year, and those who have already been hospitalized for a long period of time, and in various other conditions. However, the estimates in this paper assume that the composition of hospitalized patients does not change significantly over time. If the composition of the patient population changes significantly over time by disease, the explained variables such as the probability of care will be affected, and the coefficient values will be subject to estimation bias. However, since disease-specific characteristics are eliminated by disease-specific effects in panel fixed effects estimation, the estimated coefficient values capture the average change over time for a given disease. Therefore, if the patient composition of the disease changes significantly over time, this may be a problem, but the problem is controlled to some extent by introducing year-specific effects into the analytical model.

Variables and Data

The variables used in the analytical model are explained in more detail in this section. First, the explained variables, i.e., probability of care, average length of hospital stay, and ratio of hospitalized patients, are all measured from the patient survey. Until the 2005 survey, the Patient Survey had been conducted on the physical and mental status of inpatients in long-term care facilities. The Patient Survey is conducted every three years, and in 1999, 2002, and 2005, the results were "independent", "needs supervision (including instructions from caregivers)", "needs partial assistance", and "needs full assistance" for transferring, eating, swallowing, and cleaning up after defecation, respectively (for swallowing, the results were "able", "needs supervision", "needs full assistance", and "needs to be cleaned up after defecation"). (For swallowing, "able," "needs supervision (including caregiver's instruction)," and "unable"). Therefore, in this paper, we excluded patients who could perform these activities independently and judged that other patients needed some kind of nursing care or supervision, and measured the nursing care probability by dividing this number of hospitalized patients by the total number of hospitalized patients. The nursing care probability was calculated by age group for patients 65 years and older and those under 65 years (0-64 years). The patient survey data were constructed in such a way that they could be partially connected to the International Classification of Diseases. Therefore, by connecting with the International Classification of Diseases, we were able to construct panel data on the probability of care for about 40 diseases (see Appendix Table 1). Next, the average length of stay was obtained from a patient survey of the average length of stay for patients discharged from medical care beds and other facilities. The ratio of hospitalized patients was calculated by dividing the number of hospitalized patients in care-type beds, etc. by the number of hospitalized patients in all beds. These explained variables were also constructed as panel data in about 40 to 50 disease fields according to the International Classification of Diseases. Although data on explained variables other than the probability of care were available for periods other than 1999, 2002, and 2005, data for the same periods were used for the purpose of comparison of analysis results.

The total number of patients hospitalized for long-term care in 1999 was about 274,000, and the number increased to 325,000 in 2002 and 360,000 in 2005. This trend is similar to the aging rate and the rate of increase in the population aged 65 and over, as seen in the census, and the increase in the number of inpatients is expected to be due to this increase in the elderly population rather than a change in the system. In fact, during the period analyzed, the Medical Care Act was revised in 2001 to begin the reorganization of medical care beds, but the same standards for staffing and structural facilities were established as before so that medical care beds in hospitals and clinics could provide a medical care environment appropriate for long-term care, and there were no major changes in the medical care delivery system for medical care beds7). No major changes were made to the medical care delivery system for convalescent care ^beds7). In addition, on December 1, 2005, the medical care system reform outline by the government and the ruling party's medical care reform council stated that the reorganization of medical care beds was one of the measures to shorten the average length of hospital stay for hospitalized patients. However, these systemic changes are outside the scope of this paper's analysis period and are not considered to have a significant impact on this ^analysis8).

Next, we discuss the explanatory variables. First, the most important variables are the importance of drug stock and science. The creation of these variables and the details of the data are explained in Policy Research Institute News No. 45 and No. 55, so they will be briefly discussed ^here9). To account for the innovative nature of drugs, we identified drug components with a high contribution of science (at least one or more paper citation was found) from the scientific paper citation information of the group of patents protecting the drug. In this paper, we refer to such pharmaceuticals as science-intensive pharmaceuticals. Pharmaceuticals are formulated with a significant number of components in a single disease area, many of which also include therapeutic aids such as nutritional supplements. While these drugs may contribute to supporting treatment, they can be noise in terms of their contribution to curing the disease itself. Therefore, by looking at the efficacy of drugs with high sales by International Classification of Diseases, we have identified the main prescription components (not nutritional supplements, aids, etc.) in the disease area. Through these procedures, we constructed four drug stock variables: science-intensive main ingredient stock, science-non-intensive main ingredient stock, science-intensive non-main ingredient stock, and science-non-intensive non-main ingredient stock, by international disease category, for the period 1995-2012. Of these, the science-intensive main ingredient stock is expected to contribute the most to ^treatment10). However, these stock variables are only component-based count data, which measure the diversity of available medicines in each disease, but do not assess how important the citing scientific articles are. Therefore, we measured the importance of science based on the average number of citations (log) of scientific papers cited by pharmaceutical ingredients. This science importance index was also measured by International Classification of Diseases over the same period of time, and divided into main and non-main components.

Various medical practices other than pharmaceuticals will also affect the physical and mental status of hospitalized patients. By controlling for the implementation of these medical practices in the analytical model, the contribution of pharmaceuticals can be more accurately verified. Therefore, the surgical procedure implementation rate (the number of surgeries performed for each disease divided by the total number of hospitalized patients with the disease) by International Classification of Diseases was obtained from the patient survey and incorporated into the analytical model. In addition, since the condition of hospitalized patients is expected to depend on the age of patients, the average age of hospitalized patients by International Classification of Diseases was calculated and included in the analysis model. However, since data for these two explanatory variables were not available for inpatients admitted to convalescent care beds, data for inpatients admitted to all hospital beds were used to create the model. Since the explained variables were created only for inpatients in care facilities, there may be noise in the form of measurement error in these two explanatory variables. Therefore, we also conducted an estimation that removed them to check the robustness of the results, but there was no significant effect on the coefficient values of the importance of drug stock and science.

Analysis Results

The detailed estimation results of the analytical models (1) through (3) are summarized in the Appendix. In this section, we discuss the main results and summarize their impact based on the estimated coefficient values for the importance of drug stocks and science.

First, looking at the results of the estimation of the probability of care in equation (1) for hospitalized patients aged 65 and older (Supplementary Table 2), the coefficient values of the science-intensive main component stock and the importance of science (main component) are negative and significant for transferring and cleaning up after defecation. For example, in Model 2 for transferring, the coefficient value of the science-intensive main component stock is -0.477, which means that an increase of one science-intensive main component stock for a disease decreases the average probability of transferring care by 0.477% for that disease. The coefficient value for the importance of science (main component) is -11.14, which means that a 1% increase in the importance of science (main component) for a disease decreases the average transfer care probability for that disease by 0.111%. The results for transferring and cleaning up after a bowel movement indicate that the magnitude and significance level of the coefficient values of the explanatory variables change due to multicollinearity when the explanatory variables are swapped stepwise, but the science-intensive main component stock and science importance (main component) contribute to the decrease in the probability of caring for those patients. On the one hand, the results for the dietary intake and swallowing are similar. On the other hand, no significant effect of drug stock was found for food intake and swallowing. In these activities, the age of the hospitalized patient seems to be a significant factor. These estimation results are similar for inpatients under 65 years of age (Supplementary Table 3). However, the estimated results for patients under 65 years of age are larger in the coefficient values of the science-intensive main component stock and the importance of science (main component), suggesting that the contribution of drugs and science is ^larger11).

Figure 1 shows the measured and hypothetical values for the care probability of transfer based on the coefficient values obtained from Model 2 in Complementary Table 2. Here, the measured value is the cross-sectional average across diseases of the actually observed care probability of transfer in each of the periods 1999, 2002, and 2005. The hypothetical value is the cross-sectional mean value of the caregiving probability of transfers when both the science-intensive main component stock and the science importance (main component) variables are hypothetically unchanged from their initial values as of 1999. As can be seen from the measured values, the caregiving probability of transfers during this period has been increasing, from 58.07% in 1999 to 73.03% in 2005, an increase of 14.96%. The reason for this is not certain, as it is outside the scope of this paper's examination, but the aging rate and the increase in the population over 65 years of age may have had an impact. As for the hypothetical value, it increased to 76.82% in 2005, an increase of 18.75% from 1995. Taken together, the increase in the science-intensive main component stock and science importance (main component) during this period contributed to a 3.79% decrease in the probability of care, evaluated as of 2005 with 1999 as the base year. This 3.79% represents about 20% of the overall increase of 18.75%.

Next, the results of the estimation of the average length of stay in hospital in equation (2) (Supplementary Table 4) show that the coefficient values for the science-intensive main component stock and the importance of science (main component) are negative and significant. For example, in Model 1, the coefficient value for the science-intensive main component stock is -0.042, meaning that an increase of one science-intensive main component stock for a given disease decreases the average length of stay by 4.2% for that disease. In Model 2, the coefficient value for the importance of science (main component) is -0.970, which means that a 1% increase in the importance of science (main component) for a given disease decreases the average length of stay by 0.970% for that disease. （Similar to the result of equation (1), when the explanatory variables are gradually replaced, the magnitude and significance level of the coefficient values of the explanatory variables change due to multicollinearity, but the science-intensive main component stock and science importance (main component) can be considered to contribute to the decrease in the average length of stay. The estimated results for the average length of hospital stay are consistent with the results for transfer of care probability and cleaning up after defecation in equation (1).

Furthermore, the results of the estimation of the ratio of inpatient admissions in equation (3) (see Appendix Table 5) show no statistically significant effects of any of the explanatory variables. In Models 1 through 3, the coefficient values for science-intensive main component stock and science importance (main component) are negative but not statistically significant. In Model 4, their coefficient values are also positive but not significant due to multicollinearity. Since the expansion of the new drug stock is expected to reduce not only the number of inpatients admitted to care and other types of beds but also the number of other inpatients, the effect may not be measured by the ratio.

Finally, we report a summary of the results of two analyses conducted on the robustness of the estimation results. First, to account for the lag in drug dissemination, we also estimated the explanatory variables in a lagged model. Even if a new drug is launched on the market, it is assumed that it takes time for its effectiveness to be recognized and that it gradually gains market share. Therefore, to account for such a diffusion lag, we estimated the explanatory variables for drug stock and science importance with a one- to two-year lag as in Appendix Tables 2-5 (for example, for the probability of care in 1999, we used the number of science-intensive main ingredient stocks in 1998 or 1997 in the regression analysis). The results show that the estimation results for the 1- to 2-year lags are similar to those of the lag-free model, but the 1-year lag is the largest in terms of the magnitude of the coefficient values. Based on these results, we can say that the results are robust even for the lagged explanatory variable model (however, all explanatory variables are no longer significant in the lagged model over 3 years, so we can assume that the approximate new drug has become widespread within 1 to 2 years).

Second, although we are conducting our analysis using short-term panel data due to data limitations, it remains to be seen whether other outcome measures would also yield significant results in such a short-term analysis. For this purpose, we conducted an estimation using data on cure rates and average length of stay for inpatients in all hospital beds. The results in Tables 2 to 4 in the Appendix are for inpatients admitted to care-type beds, etc. We also tested whether the importance of drug stock and science has a similar effect on the cure rate and average length of stay for inpatients in all hospital beds during the same time period. Although detailed estimation results are omitted, we confirmed that science-intensive main component stock and science importance (main component) have a significant positive impact on inpatient cure rates, and science-intensive main component stock has a significant negative impact on average length of hospital stay. Taken together, these results are consistent with Supplementary Tables 2-4.

Conclusion

This paper analyzes new drugs and nursing care probabilities by constructing a panel data set of nursing care-disease-drug stocks by connecting patient survey and drug stock data by International Classification of Diseases for hospitalized patients in convalescent and other types of beds. The results of the estimation by the panel fixed effects estimation of diseases showed that for the physical and mental conditions of hospitalized patients, the expansion of the new drug stock utilizing the knowledge of important science had the effect of reducing the probability of needing care for them in transferring and cleaning up after defecation. For example, for the probability of care for transfers, it contributed to a 3.79% reduction in the probability of care, evaluated as of 2005, based on 1999. This 3.79% represents about 20% of the overall 18.75% increase in care probability. If such a reduction in the probability of care promotes independence of hospitalized patients, it will contribute to a reduction in the burden of care and medical care. Independence would also improve the quality of life of hospitalized patients. Thus, it is suggested that new drugs contribute to the prevention of serious illness and prolonged hospitalization of hospitalized patients, as well as to the reduction of the burden on patients and health care workers who are already hospitalized for nursing care. This in turn is expected to lead to earlier discharge from the hospital, which would also reduce the average length of stay for patients discharged from care-type beds and other facilities. The estimates in this paper are consistent with these results.

There are many remaining issues in this paper. First, regarding data limitations, only 1999, 2002, and 2005 data from the patient survey are available for this analysis, which essentially requires more up-to-date information and an analysis that takes into account recent trends in new drugs. Unfortunately, the Patient Survey no longer includes data on the physical and mental status of hospitalized patients from the 2008 survey due to the reduced burden on respondents. In addition, although the National Survey of Family Life measures the number of patients by the level of care required, there is no data by International Classification of Diseases, so the connection to the stock of pharmaceuticals used in this paper is also not possible. Thus, there are data limitations in analyzing the impact of pharmaceuticals on long-term care. As pointed out in Policy Research Institute News No. 65, "Health Status of the Elderly from the Perspective of Long-Term Care Data: Analysis Using Complementary Indicators of Healthy Life Expectancy," the linkage of disease receipt data from the National Database (NDB) and information on care requirements and care receipts from the long-term care database has been considered in recent years. In the future, it will be necessary to develop a database that is easier to use for quantitative analysis.

Next, in the estimation of the ratio of the number of inpatients admitted to medical care beds, etc. to the total number of inpatients in equation (3), no contribution of the importance of drug stock or science was found. Although sample size may be an issue, it is worth considering changing the explained variable to the ratio of the number of patients admitted to care-type beds, etc., who are care to the total number of patients admitted to the hospital. In this way, the contribution of new drugs may become clearer by looking at whether they are effective in reducing the probability of patients becoming more seriously hospitalized. Finally, a disease-specific analysis is needed. Although the probability of care varies widely by disease, this paper removes this effect as a disease-specific effect and does not estimate the contribution of new drugs by disease. However, a disease-specific analysis would require expanded data over time for each disease, and data from patient surveys would limit the analysis. For example, in the U.S., there is a database that tracks drug administration and subsequent health status at the patient level, which enables disease-specific analysis while also taking patient attributes into ^account12). Future efforts should be made to develop such a database as well as to improve the analytical model.

Return to News List

TOP