The health of elderly patients is typically characterized by more than one disorder [1]. More than 10% of all Austrians aged >50 years accumulate more than 10 different diagnoses over a period of 2 years [2]. Treatment of such highly multimorbid patients often requires the involvement of many different care providers [3,4] taking age-specific and sex-specific differences in physiology and health care–seeking behavior into account [5]. Yet, most health care systems are still configured to treat individual diseases rather than individual multimorbid patients [6]. It is therefore an open challenge to ensure sufficient care coordination among different types of health care providers to adequately treat an aging population [7]. Most findings on long-term readmission risk so far have had an isolated focus on single diseases — for instance, pneumonia [8], colorectal surgery [9], depression [10], or chronic obstructive pulmonary disorder (COPD) [11] — and take only a few predictor variables (eg, medical history) into account [12]. In addition, many studies focus on short-term (30-day or 90-day) readmissions, whereas studies of longer-term risks for patients with chronic complex disorders such as diabetes remain underrepresented in the literature [13].

Digitalization in the health sector has led to increasing availability of observational health care data like electronic health records or medical claims data [14]. The emerging field of network medicine [15,16] strives to leverage such data to improve our understanding of multimorbidity [17] and how care providers coordinate themselves in the treatment of such patients [18]. Complex multimorbid health states of patients can be conceptualized by means of networks (collections of nodes connected by links) in which diseases are nodes that are linked if they tend to co-occur in patients. These comorbidity networks can be used to predict future changes in health as patients are most likely to acquire diseases in close network-proximity to those that they already have [2,19,20]. Networks of care providers have been studied through the analysis of patient-sharing relations [21]. In such patient-sharing networks, providers are represented as nodes connected by links that indicate patient flow between them [18]. It has been shown that the structure of such networks can be related to variations in treatment outcomes [18], cost and intensity of care [22,23], as well as spending for and utilization of health services [24,25].

In this work, we quantified for the first time the long-term readmission risk for 97 frequent diagnoses (ICD-10 3-digit codes) associated with the first hospital stay as a function of age, sex, and the involvement of 13 different types of medical specialists. We propose a novel network statistical modelling approach illustrated in Figure 1. Using an administrative database containing data for almost 2 million patients, we identified all patients aged >50 years with at least one hospital admission (index hospitalization) and followed them for >3 years. There are 4 different types of trajectories that a patient can take after first admission (index hospitalization): (1) no second admission and no specialist contact over the next 3 years (see patient 1 in Figure 1A), (2) one (or several) specialist contact but no second admission (see patient 2 in Figure 1B), (3) second admission without specialist contact (see patient 3 in Figure 1C), or (4) second admission with specialist contact (see patient 4 in Figure 1D). By considering patients with the same diagnosis from the first admission (index diagnosis) and adjusting for age, we can then estimate separately for men and women how contact with a specific type of provider changes the readmission risk for any combination of index diagnosis, readmission diagnosis, and type of specialty.

Descriptive statistics of the study population are shown in Table 1. The study population was skewed toward the female sex (130,968/225,238, 58%) with average ages (taken at the beginning of the observation window) of 65 years (men) and 68 years (women). With SDs of 9.7 years (men) and 11 years (women), both sexes had similar and rather broad age distributions. Our inclusion criteria resulted in 97 diagnoses (ICD-10 codes) and 13 different types of specialists. Average numbers of diagnoses and types of specialists involved in the treatment were similar between men and women.

In this work, we present a comprehensive analysis of sex-specific long-term readmission risks (measured from 90 days to 3 years after the index hospitalization) where we systematically tested how contact with 13 medical specialties impacts readmission risks for 97 diagnoses associated with the first stay. The network visualization reveals that our analysis is indeed based on meaningful flows of patients between different care settings. For instance, we found a dominant flow from lab testing to senile cataract consistent with the fact that such testing is often performed in preoperative screenings to detect risk factors for complications such as diabetes. There are multiple meaningful flows from radiology to musculoskeletal diseases, a link from dermatovenereal diseases to skin cancer, or from ENT specialists to nontoxic goiter. In all cases, our results mean that patients that had contact with a specialist showed a tendency later for reduced readmission risks for the given diagnoses compared to patients without such contact (ie, the links in the network do not just show frequent flows of patient but specialty-diagnosis combinations that contribute to the observed reductions in readmission risks). Other links were less clear. For instance, we found a tendency that contacts with psychiatry reduce readmission risks for pneumonia. Recent epidemiological findings suggest that depression is indeed a risk factor for hospitalization due to pneumonia [30] and that psychological distress is related with a higher risk of pneumonia [31]. Furthermore, lifestyle factors (eg, substance abuse), psychiatric conditions (patients’ compromised ability to recognize health problems) as well as side effects of antipsychotics (worsened respiratory muscle functioning) might cause this association [32]. Our results could therefore indicate that contact with psychiatric specialists mitigate these risk factors and thereby reduce pneumonia-related readmissions.

Overall, we found the largest readmission risks after hospital stays associated with chronic complex diseases for which high readmission rates have already been described in the literature. These diseases include various types of cancer including rectal cancer, with a 30-day readmission rate of 10.1% [33], and lung cancer, with a 30-day readmission rate of 13% and 90-day rate of 22% [34]. Diabetic and other retinopathies often occur with type 2 diabetes as the main diagnosis, for which 30-day readmission rates are 8.5%-13.5% [13]. For chronic kidney disease, the 90-day readmission rate has been estimated at 11.7% [35]. These risks cannot be directly compared to the long-term readmission risk (where we exclude readmissions within the first 90 days) considered in our study.

The involvement of medical specialists reduces the need for long-term readmissions by up to 50% depending on the index diagnosis. Chronic complex diseases are among those for which we observe the strongest reductions in readmission risk after contact with medical specialists. Our observation of the greatest reduction for patients with acute myocardial infarction is in line with findings of reduced mortality (up to 19% over an 18-month follow-up) for patients with myocardial infarct who receive follow-up care by cardiologists and internists when compared to patients without such contact [36]. The second greatest reduction was observed for diabetic and other retinopathies (H35), which often occurred with type 2 diabetes as the main diagnosis. These findings are in line with reports that a lack of postdischarge outpatient visits in patients with diabetes is one of the strongest risk factors for short-term (30-day) readmissions [37] and that postdischarge office visits to adjust the diabetes regimen contribute to a decreased risk of short-term readmission [38]. While there is mixed evidence to which extent poor glycemic control is also a risk factor for longer-term readmission risk [13], our findings clearly show that specialist care after discharge is related to a strongly significant reduction in readmission risks of down to 62% (men) and 60% (women) compared to patients without such contacts. Similar diabetes-related observations might be relevant for patients with hypercholesterolemia and hyperlipidemia (E78), which are frequent diabetic comorbidities, who all showed significant reductions in readmission risk. We found that the contact related with reductions in readmission risk for diabetes patients was concentrated on visits at outpatient wards, internal medicine, and radiology, among other specialists. Diabetes is indeed a complex disease requiring the involvement of multiple types of health care providers. Treatment should take place in strict agreement with the corresponding guidelines, including quarterly physician visits and a high continuity of care, to minimize the risk for diabetic complications.

For COPD patients, it has been observed that the involvement of physiotherapists and various pulmonary and respiratory specialists can reduce readmission risks, which is consistent with our finding of a strongly reduced readmission risk for patients with COPD [39]. Finally, in relation to our results for the relative readmission risk for ischemic heart disease, it was reported that patients had significantly lower 60-day readmission rates when they were treated by multiple providers, including surgeons and nonsurgeons [40].

Men and women had comparable probabilities of contacting different types of medical specialists after the index admission; there was no significant difference in how likely men and women seek specialists. A Swedish register study found that most of the sex differences in health care consumption can indeed be explained by an increased level of reproduction-associated care (not considered in our work) and women’s higher share in mental and behavioral disorders and diseases of the musculoskeletal system [41]. We found contact probabilities that range from around 8% (psychiatry) to more than 56% (radiology). We did not include primary care providers (eg, general practitioners) in this analysis as almost everyone in the study population had such contacts; therefore, their contact probabilities were close to 100%. After contact with specialists, we observed significantly reduced readmission risks (risk reductions of up to 50%) for almost all specialties, including lab testing, radiology, psychiatry, dermatovenereal diseases, and ENT specialists, whereas pulmonary disease specialists and orthopedics show a rather risk-neutral profile. These findings might reflect that follow-up by specialists generally means more tailored risk detection and improved disease management, but also that patients seeking care from specialists might be more engaged and vigilant compared to those patients that do not seek specialist care. In the present form, our analysis does not allow us to disentangle these effects of targeted prevention and health care–seeking patient behavior.

In terms of sex differences, we found that men overall have higher readmission risks than women. While the raw readmission frequencies were similar for men and women (Table 1), the diagnosis-specific analysis clearly revealed that men have increased readmission risks after adjusting for age and pre-existing condition (index diagnosis). Our findings also showed that the difference between men and women in readmission risks are not due to differences in how likely they are to seek contact with a specialist. In the following, we give two plausible mechanisms that could in principle contribute to the observed sex biases (or lack thereof). First, men might utilize the health care system only at more severe stages of disease compared with women, therefore also showing higher readmission risks. Second, it could be that women are more compliant when consulting specialists and therefore show better outcomes (ie, reduced readmission risks). However, the second explanation is at variance with our result that, after having had a specific type of contact, men and women show similar reductions in readmission risk. The assumption that these sex biases are indeed due to differences in utilization is further corroborated by findings of delayed health-seeking behavior in men compared with women [42].

Our work has several limitations that mostly relate to the administrative dataset used. We have no information on which kind of procedures were performed during the admission and contact with a specialist and no knowledge on results from medical tests other than the diagnosis codes. We cannot guarantee that we indeed observed all admissions of the study population, especially since the data cover only a region of Austria. However, as this bias should have similar effects on index hospitalizations and readmissions, as well as men and women, such coverage issues should only have a limited impact on comparisons of readmission risks. Similar biases might influence the probabilities of contact with specialists. We only considered whether at least one contact took place, but not the specific number of contacts.

To conclude, our results emphasize that specialist aftercare can provide a strong contribution to the reduction in long-term rehospitalization. These effects vary substantially across diagnoses and are most pronounced for outcomes such as myocardial infarction where specialist treatment has already been shown to improve survival. While we find tentative evidence for delayed health-seeking behavior towards medical specialists in men when compared to women, both sexes show similar levels of readmission risk reduction after specialist care. These sex biases require further research into their physiological, biological, social, and psychological causative processes.