Counties and county-level cities in China's central (e.g. Dingyuan in Anhui Province) and western (e.g., Huining and Weiyuan in Gansu Province and Yilong in Sichuan Province) regions were designated as sample areas. The largest and most capable public comprehensive hospital in each sample county was selected as a sample hospital. The reimbursement and payment levels of the primary medical insurance, which may be associated with medical costs, as well as hospitalization behaviors (10), in the four counties are similar.

Retrospective study was conducted to evaluate the appropriateness of admission. To study the changes in the appropriateness of admission, the evaluation of medical records is divided into two parts. One is to evaluate the appropriateness of admission according to the “admission records” in medical records. Admission records is a part of medical records, including identification, chief complaint, history of present illness, past history, family history, marital history, general physical exam, etc., which can reflect the patient's condition at the time of admission. The second is according the progress note, auxiliary examinations, doctor's orders and prescriptions, etc., in the medical records other than the “admission records” to evaluated the appropriateness of admissions which were defined as “inappropriate” in first part evaluation.

The progress note, auxiliary examinations, doctor's orders and prescriptions, etc., can reflect the development of the disease. Combined with the evaluation results of the two parts to explore the changes of the appropriateness of admission (shown in Figure 1).

In accordance with the existing research (11), the estimated inappropriate admission rate P is 16%, relative tolerance δ is 0.07, absolute tolerance d is 0.07^*16% = 1.12%, the significance level α=0.05, and the one-sided standard normal deviation Zα =1.96. The equation of sample size (N) is as follows:

Considering the quality of medical records, 1,200 medical records in 2017 were planned to be collected from each hospital. Firstly, admission for delivery is necessary and appropriate, and the services required for these cases are reasonable, so these records in obstetrics department were excluded considering the pertinence of AEP. Then, the corresponding quantity of medical records was selected from the remaining medical service departments in accordance with the proportion of patients in the department accounted for the total quantity of patients in all departments. After excluding medical records that had too many missing values and serious logic error, 4,343 medical records were taken into analysis (shown in Figure 2).

AEP was originally developed by Gertman and Restuccia of the medical research center at Boston University school of medicine (18). A total of 16 criteria were used to evaluate the appropriateness of admission, including severity of illness and service intensity. Based on the AEP criteria of the United States, many countries have explored different AEP criteria that meet their own conditions (19, 20). The AEP criteria for county hospitals in China has been developed in 2014 (Supplementary Table for details) (10).

All medical records were evaluated by AEP criteria for county hospitals in this study. The medical records were evaluated by two trained experts. These highly trained experts involved in health service and policy research more than 5 years, all of whom received Ph.D. degrees and committed to making fair judgments on records in a strictly standard manner. They assessed the appropriateness of each medical records independently. The admission was considered appropriate if an actual value in medical record corresponded to the standard value in AEP. If no values in medical record conform to the AEP criteria, the admission was inappropriate. In other words, if any single one of the criteria was met, it would indicate that admission was appropriate. If no one of the criteria was met, would indicate that admission was inappropriate. Therefore, if a case was judged to be appropriate on the basis of admission records, the result of the admission was appropriate.

Each medical record was evaluated by each expert on the basis of the admission records and the rest part of medical records. The records that had different evaluation results between the two experts were judged by a third party, usually clinical experts. At last, reasonable evaluated results for admission record and the rest part of medical records were chosen after a comprehensive evaluation. Then, comparing two results reveals changes in admission appropriateness.

The diseases were categorized by using the International Classification of Diseases 10th revision (ICD-10) and were classified into several common disease categories. Age was divided into five groups according to children (0–14 years old), adolescents (15–19 years old), young adults (20–39 years old), the middle-aged (40–59 years old) and the aged (more than 59 years old). The medical services were medical, surgical, gynaecologic and pediatric. The severity of the patient upon admission, which were classified as “general, urgent, serious, and dangerous”.

The results of the admission appropriateness evaluation judged by the two parts were compared by McNemar's test. The binary logistic regression model was used to identify the determinants of changes in admission appropriateness. The dependent variable in the regression model was a binary variable, that is, “changed to appropriate or not” (“Yes” = 1, “No” = 0). The independent variables were first identified based on previous studies and whether could be obtained from the medical records, which included: Gender, Age, Type of medical insurance, Frequency of hospitalization, Medical service, Severity of the patient upon admission, History of disease, Having more than one disease, Disease category, With chronic disease. Then forward stepwise method was used to select independent variables in the final regression model (21).

The regression model is as follows:

The statistical analysis was performed by IBM SPSS Statistics 20.0. P < 0.05 was considered statistically significant.

As shown in Table 1, by evaluating admission records, 942 (21.69%) records were appropriate, 3,401 (78.31%) records were inappropriate. Judging the 3,401 records again according to the rest part of the medical records, it was concluded that 2,224 of the 3,401 inappropriate admissions changed to appropriate, accounting for 65.39%. 1,177 cases were still inappropriate, accounting for 34.61%. A total of 3,166 records were judged as appropriate admissions, and 1,177 records were judged as inappropriate admissions. Inappropriate admission rate was 27.1%. The difference between the two parts evaluation was statistically significant (P < 0.001).

As shown in Table 2, 65.39% of the 3,401 inappropriate admissions changed to appropriate finally. The appropriate rate of females (67.1%, P = 0.043) was higher than that of males (63.8%, P = 0.043). The highest appropriate rate was found among the older adult (74.4%, P < 0.001), whereas the lowest was found among children (45.3%, P < 0.001). Inpatients with “medical assistance” had higher appropriate rate (84.8%, P < 0.001) than inpatients with new rural cooperative medical scheme (NRCMS) and “medical insurance for urban residents and workers”. Inpatients in Medical (69.1%, P < 0.001) and Gynaecologic (72.8%, P < 0.001) had high appropriate rate. Inpatients who had “serious” status upon admission (70.2%, P < 0.001), who had more than one disease (76.8%, P = 0.006) and who with chronic diseases (74.1%, P < 0.001) were prone to get appropriate finally. Inpatients suffering from urinary diseases (81.9%, P < 0.001), genital diseases (80.3%, P < 0.001) and circulatory diseases (77.3%, P < 0.001) had the highest appropriate rate, whereas those suffering from respiratory diseases had the lowest (44.2%, P < 0.001).

As shown in Table 3, binary logistic regression analysis showed that age, type of medical insurance, medical service, severity of the patient upon admission and disease category were determinants of affecting the changes in the appropriateness of admission in county hospitals. Older patients (OR = 3.658, 95% CI: [2.462–5.435]; P < 0.001) were positively associated with changing from “inappropriate” to “appropriate” than younger patients. Inpatients covered by medical assistance (OR = 2.661, 95% CI: [1.926–3.676]; P < 0.001) were positively associated with being considered appropriate admissions at discharge than those covered by NRCMS. Self-payment (OR = 0.656, 95% CI: [0.434–0.993]; P = 0.046) had the lowest possibility be evaluated as appropriate at discharge. Compared with others, inpatients in the pediatric were positively associated with changing from “inappropriate” to “appropriate”. In terms of severity of the patient upon admission, urgent patients (OR = 1.392, 95% CI: [1.109–1.747]; P = 0.004) and serious patients (OR = 1.48, 95% CI: [1.041–2.105]; P = 0.029) were positively associated with changing from “inappropriate” to “appropriate” than ‘general' patients. They rarely occurred among the patients labeled as “dangerous” (OR = 0.593, 95% CI: [0.421–0.835]; P = 0.003). Compared with circulatory diseases, the case be evaluated as “appropriate” at discharge was positively associated with the urinary diseases (OR = 1.709, 95% CI: [1.019–2.865]; P = 0.042) and genital diseases (OR = 2.998, 95% CI: [1.737–5.174]; P < 0.001), whereas the opposite finding was observed for respiratory diseases (OR = 0.347, 95% CI: [0.268–0.451]; P < 0.001) and skeletal and muscular diseases (OR = 0.556, 95% CI: [0.355–0.873]; P = 0.011).

As so far, this study is the first one to concern changes in the appropriateness of admission after the patients was admitted in county hospitals in rural China. The study showed that 65.39% of the inappropriate cases at admission changed to appropriate at discharge. The result may be attributed to the following reasons. First, the admission appropriateness evaluation when the patients is admitted is based on the patients' indications before and at the time of admission. However, diseases are complex and have insidious characteristics, and change constantly (22). In addition, different patients have different symptoms, severity and different development stages of diseases at the time of admission. Some patients are in the early stage of disease and their disease characteristics does not fully emerge. In other words, the patients may not meet the admission criteria at the time of admission but their conditions might get serious after admission. The risk of deterioration and vulnerabilities of some patients increase the uncertainty of their conditions (23). All of this make the admission appropriateness evaluation when patients were admitted less comprehensive and accurate. Second, as an evaluation instrument, the AEP cannot fully substitute for the professional judgments of clinicians (24), which results in the appropriateness assessment of admissions lack of flexibility (18, 25).

The study found that older patients' admissions were more likely to go from “inappropriate” to “appropriate” than younger patients. This finding may be attributed to older patients have multiple types of diseases occurring simultaneously, majority of which are chronic diseases. Insidious onset and clinical symptoms are not typical but start to emerge after admission (26, 27). However, younger patients are mainly affected by acute diseases, which are easily identified (28, 29). Given that the middle-aged and older adult patients constitute the majority (69.7%) in medical assistance, inpatients are easily judged to be appropriate admissions at discharge compared with the NRCMS inpatients (30, 31). The admission of patients with medical insurance for urban residents is not easy change to appropriate, indicating that compared with rural inpatients, urban residents are prone to use hospital services unreasonably, while urban workers are on the contrary (32, 33).

As one of the special groups, children had different characteristics compared with adults. The function of children's various organs will develop with age. For the same pathogenic factor, there are considerable differences in the pathological reaction and disease development between children and adults (34). For example, pneumonia caused by pneumococcus is more common in infants with bronchopneumonia, while lobar pneumonia may be present in adults and older children (35, 36). The types and clinical manifestations of childhood diseases are quite different from that of adults. Therefore, there may be deviation in using the same set of criteria to measure the admission appropriateness for special population. Inpatients in the pediatric are easily categorized as inappropriate in admission but they were appropriate admission actually. This suggests that the admission recognition criteria applicable to general patients may not be suitable for pediatrics. It is necessary to add key indicators that meet the characteristics of pediatric patients and further demonstrate and test in practice.

This study showed that patients with “dangerous” status upon admission are not likely to go from inappropriate to appropriate. Patients with “dangerous” status upon admission are more likely to reach the admission criteria than patients with general status theoretically. Patients who are admitted in a “dangerous” condition are easily rated as appropriate admission. The “dangerous” cases were rated as inappropriate may be related to doctors' judgment bias at the time of consultation and the nonstandard writing of medical records. The circulatory disease, namely cardiovascular and cerebrovascular diseases have very complex symptoms and conditions (37, 38). Due to the diversity of symptoms and the logicality of the patient's description varies greatly, the development degree of the disease cannot be clearly determined by the self-reported or other-mentioned symptoms nor the “sight, touch, knock and listen” of physical examination (39–41). The criteria of pulse and blood pressure in AEP cannot cover the signs of circulatory disease. The patients who are at risk for underlying diseases were more likely to reach the admission criteria at discharge. The urinary and genital diseases are mostly caused by infection because of the particularity of the diseases themselves (42). Mycoplasma is one of the pathogens causing urinary and genital infections. However, due to its small size, it grows slowly in the medium and is difficult to be observed, so the clinical diagnosis of mycoplasma infection is more difficult (43, 44). So, it is difficult to determine the severity of the urinary and genital diseases at the time of admission accurately. Nevertheless, respiratory diseases can be diagnosed by listening to breathing sounds, and the skeletal and muscular diseases can be easily diagnosed through viewing and touching (45, 46). The external characteristics of these two diseases are relatively obvious, and it is very easy to be diagnosed at the time of admission. The probability of going from inappropriate to appropriate is lower compared to circulatory, urinary and genital diseases.