Fig 1.
Disease portfolio development.
Illustration of disease portfolio trajectories following heart disease diagnosis for three individuals.
Table 1.
Overview of explanatory variables in the statistical analysis.
Table 2.
Population characteristics at time of HD diagnosis according to sex.
Table 3.
Diagnosis prevalence of HD population according to sex.
Prevalences are reported at time of HD diagnosis, as well as for the entire span of the observed disease trajectories.
Table 4.
Top 10 disease portfolio prevalence at time of HD diagnosis according to sex.
Fig 2.
Diagnosis counts by age for the population of Danish chronic heart disease diagnosed individuals in the period 1995–2015.
The figure shows males (left) and females (right). A diagnosis is counted when the individual obtains the algorithmic chronic disease diagnosis. The diagnoses are ordered, so diagnoses with larger variance in counts are on top.
Table 5.
Effects for statistical significant explanatory variables on the postponement time until the next disease diagnosis.
Positive and negative effects correspond to increased and decreased diagnosis postponement times.
Fig 3.
Heatmap on importance of different variables on transition probabilities to distinct next diagnosis endpoints.
Results were obtained from a backwards selection of covariates where each unique disease portfolio was modelled separately for each next diagnosis endpoint. The importance of one indicates that the term was kept in all disease portfolios. The figure is sorted by sums of importance across endpoints (left to right) and across variables (top to bottom).
Table 6.
Most important variables for transition probabilities.
Top five significant (1%) variables with the largest discriminative power across different next diagnosis endpoints as measured by the absolute size of the coefficients in the logistic regression models. The variables are presented by the coefficient sign, where a positive (negative) coefficient corresponds to an increased (decreased) probability of the outcome relative to the other possible outcomes. A colon gives coefficients related to the presence of two simultaneous diseases (Disease 1:Disease 2).
Fig 4.
Effect of multimorbidity on postponement times.
Estimated mean postponement time of next chronic diagnosis given survival for retired males and females of no education at a set multimorbidity level (left). The mean postponement times are estimated as an average of estimated postponement times for all possible combinations of chronic conditions given the multimorbidity level, weighted by the observed frequency of these combinations. The blue shaded area represents the proportion of HD individuals at risk of a new event. For reference, the right plot represents the same estimates in a model where the new disease event and death are considered a combined event.
Fig 5.
Sex-specific disease trajectories contrasting educational attainment.
Disease trajectories starting from the most common triad of diseases for males (left) and females (right) of no education (top) and long education (bottom). Trajectories are constructed for retired individuals with calendar time and age set at the mean levels at t = 0 (70.09 years of age and 2003.37 year time, respectively). The length of the arrows corresponds to the modelled diagnosis postponement time. The width of the arrows corresponds to the modelled probability of obtaining the diagnosis next.
Fig 6.
Estimated mean postponement times by educational attainment levels.
The figure presents the mean postponement time of the next chronic diagnosis given survival for retired males (left) and females (right) at various educational levels. Estimations are made starting from the most common triad of diagnoses heart disease, hypertension, and high cholesterol (see Fig 5). The second stage postponement times are a weighted average with weights determined by the transition probabilities of the first stage diagnoses.