Final Multiple Linear Regression Research Articles

Machine-learning-based similarity meets traditional QSAR: “q-RASAR” for the enhancement of the external predictivity and detection of prediction confidence outliers in an hERG toxicity dataset

Recently, the concept of quantitative Read-Across Structure-Activity Relationship (q-RASAR) has been introduced by using various Machine Learning (ML) - derived similarity functions in the traditional quantitative structure-activity relationship (QSAR) modeling framework with the objective of enhancing the external predictivity of models while using the same available chemical information content. The present study uses the hERG K+ channel inhibition cardiotoxicity, a pharmaceutically relevant endpoint, as the modeling set for making predictions using the novel q-RASAR approach, as the approach combines the merits of QSAR and Read-Across, and generates simple and interpretable models using various similarity and error-based measures as descriptors. The cardiotoxicity data (in terms of pIC50 values) were collected from the literature. The curated data set was then divided into training and test sets using the sorted response-based division algorithm. The important set of features was identified based on the internal validation metrics of initial genetic algorithm models. Based on the features selected in the final Multiple Linear Regression (MLR) model, RASAR descriptors were computed using a tool available from https://sites.google.com/jadavpuruniversity.in/dtc-lab-software/home. The RASAR descriptors were then merged with the previously selected features, and an MLR q-RASAR model was generated using the grid search approach. The prediction outliers were then identified using the novel DTC Applicability Domain Plot, and the q-RASAR models were used for predictions after the removal of prediction outliers. A final Partial Least Squares (PLS) q-RASAR model was generated to obviate inter-correlation among descriptors. Various other Machine Learning approaches were also employed with the optimization of relevant hyperparameters based on the cross-validation approach, and the final test set prediction results were compared. Based on the performance in the test set predictions and interpretability, the PLS q-RASAR model was chosen as the final model which provided enhanced predictivity in comparison to previously reported models even without using 3-D descriptors. This model can thus be used for the quick screening of molecules, even before their synthesis, to estimate their cardiotoxic potential thus prioritizing molecules for further experimental testing in the drug discovery pipeline. A Java-based prediction tool has also been developed for the quick screening of cardiotoxic properties of query compounds and made available from https://sites.google.com/jadavpuruniversity.in/dtc-lab-software/home.

Machine Learning-Assisted Identification and Quantification of Hydroxylated Metabolites of Polychlorinated Biphenyls in Animal Samples.

Laboratory studies of the disposition and toxicity of hydroxylated polychlorinated biphenyl (OH-PCB) metabolites are challenging because authentic analytical standards for most unknown OH-PCBs are not available. To assist with the characterization of these OH-PCBs (as methylated derivatives), we developed machine learning-based models with multiple linear regression (MLR) or random forest regression (RFR) to predict the relative retention times (RRT) and MS/MS responses of methoxylated (MeO-)PCBs on a gas chromatograph-tandem mass spectrometry system. The final MLR model estimated the retention times of MeO-PCBs with a mean absolute error of 0.55 min (n = 121). The similarity coefficients cos θ between the predicted (by RFR model) and experimental MS/MS data of MeO-PCBs were >0.95 for 92% of observations (n = 96). The levels of MeO-PCBs quantified with the predicted MS/MS response factors approximated the experimental values within a 2-fold difference for 85% of observations and 3-fold differences for all observations (n = 89). Subsequently, these model predictions were used to assist with the identification of OH-PCB 95 or OH-PCB 28 metabolites in mouse feces or liver by suggesting candidate ranking information for identifying the metabolite isomers. Thus, predicted retention and MS/MS response data can assist in identifying unknown OH-PCBs.

P-063 Multivariate analysis identifies oxidative stress levels and acrosome status of motile spermatozoa as independent predictors of motile sperm hyaluronan binding ability

Abstract Study question Is it possible to predict motile sperm hyaluronan binding ability (HBA) from motile sperm functional parameters? Summary answer Using multivariate analysis, oxidative stress levels and acrosome status of motile spermatozoa, derived from semen swim-up, were identified as predictors of sperm HBA. What is known already Semen analysis is a gold standard practice for fertility evaluation. However, motility and morphology of spermatozoa are not always indicative of the molecular makeup of sperm cells. In this context, several sperm functional tests were developed to enhance our understanding of spermatozoa biology. Interestingly, several studies have shown that the plasma membrane of mature sperm cells contain hyaluronan binding sites. HBA was associated with intact acrosome, decreased DNA damage, and high viability. Nevertheless, it has not been demonstrated yet if hyaluronan binding ability can be predicted solely by swim-up derived- motile spermatozoa functional parameters. Study design, size, duration A prospective study was performed at Al-Hadi Medical Center, beirut, Lebanon, between January and July 2020. Freshly ejaculated semen samples obtained from infertile men with concentration of ≥ 20 x 106/ml and with total sperm motility of ≥ 30% were included. The primary outcome was to assess if the motile sperm functional parameters may influence HBA. A sample size of 54 was calculated using power = 85% and alpha = 5% for multiple linear regression. Participants/materials, setting, methods 54 semen samples were assessed according to WHO guidelines (2010). Thereafter, samples were processed using the semen swim-up technique. Supernatants containing motile spermatozoa were collected for further analysis. In motile spermatozoa-enriched fractions, chromatin integrity was assessed using toluidine blue (TB) staining. Oxidative stress levels were detected using a nitroblue tetrazolium (NBT) test. Acrosome status was determined using triple staining technique. HBA was evaluated via a commercially available hyaluronan binding assay (HBA® Assay). Main results and the role of chance In the motile spermatozoa-enriched fractions, mean ± standard deviations for concentration, progressive motility, non-progressive motility and normal morphology were 26.17±11.48, 76±18.97, 20.94±12.61 and 0.75 ± 1.4, respectively. In addition, mean ± standard deviations for TB + cells, NBT + cells, living spermatozoa with intact acrosome, and HBA were 16.42 ± 13.4, 57.4 ± 9.4, 93.9 ±3.2, and 48.98±11.95, respectively. The values of the HBA had a non-normal distribution; therefore, these values were first linearized by taking the logarithm of the outcome. Afterwards, multiple linear regression (MLR) was used to predict the outcome. The overall regression of the final model was statistically significant (R2 = 0.346), F(4.43) =5.962, p = 0.001. The results of the final MLR model indicated that a unit increase in non-progressive motility (%) (p = 0.002) and NBT + cells (%) (p = 0.009) increased the HBA by 1.505 units and 1.409 units, respectively. However, a unit increase in dead spermatozoa with unreacted acrosome (%) (p = 0.032) and living spermatozoa with reacted acrosome (%) (p = 0.026) decreased the HBA by 1.317 units and 1.342 units, respectively. Limitations, reasons for caution This is an interim study and these results should be taken as preliminary. In order to confirm these findings, further studies including larger sample size and motile spermatozoa derived from semen samples with abnormal parameters (such as severe asthenozoospermia, leukocytospermia, severe oligozoospermia) are required. Wider implications of the findings Several studies indicated that using damaged spermatozoa may affect the intracytoplasmic sperm injection (ICSI) outcomes. In addition, using damaged spermatozoa during fertilization may predispose the offspring to various diseases later in life. These results may be considered as an impetus to reconsider sperm selection during ICSI. Trial registration number Not applicable

An Experimental Urban Case Study with Various Data Sources and a Model for Traffic Estimation.

A reliable estimation of the traffic state in a network is essential, as it is the input of any traffic management strategy. The idea of using the same type of sensors along large networks is not feasible; as a result, data fusion from different sources for the same location should be performed. However, the problem of estimating the traffic state alongside combining input data from multiple sensors is complex for several reasons, such as variable specifications per sensor type, different noise levels, and heterogeneous data inputs. To assess sensor accuracy and propose a fusion methodology, we organized a video measurement campaign in an urban test area in Zurich, Switzerland. The work focuses on capturing traffic conditions regarding traffic flows and travel times. The video measurements are processed (a) manually for ground truth and (b) with an algorithm for license plate recognition. Additional processing of data from established thermal imaging cameras and the Google Distance Matrix allows for evaluating the various sensors’ accuracy and robustness. Finally, we propose an estimation baseline MLR (multiple linear regression) model (5% of ground truth) that is compared to a final MLR model that fuses the 5% sample with conventional loop detector and traffic signal data. The comparison results with the ground truth demonstrate the efficiency and robustness of the proposed assessment and estimation methodology.

Impaired Lower Extremity Biomechanics, Hip External Rotation Muscle Weakness, and Proximal Femoral Morphology Predict Impaired Single-Leg Squat Performance in People With FAI Syndrome

Background: Impairments in squat depth have been reported in patients with femoroacetabular impingement syndrome (FAIS). However, little is known about single-leg squat (SLS) performance in these patients, despite this task being commonly used in the rehabilitation and training settings. Purpose/Hypothesis: The aims of this study were (1) to investigate whether patients with FAIS demonstrate differences in SLS performance compared with healthy controls and (2) to determine whether dynamic range of motion (ROM), muscle strength, hip morphologic measures, hip pain, and hip-specific function predict SLS performance in patients with FAIS. We hypothesized that patients with FAIS would demonstrate impaired SLS performance and that impaired hip biomechanics, muscle strength, and hip-specific function would predict squat performance in patients with FAIS. Study Design: Controlled laboratory study. Methods: Three-dimensional (3D) kinematic data were collected at 100 Hz using a 20-camera 3D motion capture system during 3 SLS trials in 34 patients with FAIS and 26 healthy controls. Isometric muscle strength was tested with a stationary handheld dynamometer in all participants. Squat performance was quantified by squat depth (in meters), and the biomechanical variables of dynamic ROM of the pelvis, the hip, the knee, and the ankle in all planes were calculated. In patients with FAIS, femoral and acetabular morphology were measured using radiographic alpha angles and lateral center-edge angles. Hip pain and hip-specific function were measured using the visual analog scale for pain and the Hip Outcome Score Activities of Daily Living subscale, respectively. Two-tailed independent-samples t tests were used to determine between-group differences for squat depth, dynamic ROM variables, and muscle strength. A hierarchical multiple linear regression (MLR) model was used to determine whether biomechanical variables, muscle strength, hip morphology measures, hip pain, and hip-specific function were predictors of squat depth. All statistical analyses were performed using SPSS Version 26. Results: There were no between-group differences in age (FAIS, 30.0 ± 7.0 years vs controls, 27.3 ± 7.0 years; P = .18) or body mass index (FAIS, 23.1 ± 2.8 vs controls, 22.6 ± 3.2; P = .51). Squat depth was less in patients with FAIS compared with healthy controls (FAIS, 0.24 ± 0.4 m vs controls, 0.29 ± 0.05 m; P < .001). In the sagittal plane, patients with FAIS demonstrated less dynamic ROM of the hip (FAIS, 67.8°± 12.4° vs controls, 79.2°± 12.5°; P = .001) and the knee (FAIS, 71.9°± 9.4° vs controls, 78.9°± 13.2°; P = .02) compared with controls. Patients with FAIS also demonstrated a less dynamic coronal plane pelvis ROM (FAIS, 11.3°± 5.0° vs controls, 14.4°± 6.7°; P = .044). Patients with FAIS had reduced hip muscle strength of the hip external rotator (FAIS, 1 ± 0.3 N/kg vs controls, 1.2 ± 0.3 N/kg; P = .034), hip internal rotator (FAIS, 0.8 ± 0.3 N/kg vs controls 1 ± 0.3 N/kg; P = .03), and hip flexor (FAIS, 4 ± 1.1 N/kg vs controls, 4.8 ± 1.2 N/kg; P = .013) muscle groups. The hierarchical MLR revealed that the dynamic ROM of the hip, the knee, and the pelvis, the hip external rotation muscle strength, and the femoral alpha angles were all significant predictors of squat performance, and the final MLR model explained 92.4% of the total variance in squat depth in patients with FAIS. Conclusion: Patients with FAIS demonstrate impaired SLS squat performance compared with healthy controls. This impaired squat performance is predominantly predicted by sagittal plane knee and hip biomechanics and hip external rotator strength, and less by frontal plane pelvic ROM and hip morphology in patients with FAIS. Clinical Relevance: Clinicians should focus treatment on improving dynamic ROM and hip external rotator muscle strength to improve squat performance; however, femoral morphology should also be considered in the treatment paradigm.

In Silico Modeling of Small Molecule Carboxamides as Inhibitors of SARS-CoV 3CL Protease: An Approach Towards Combating COVID-19.

The quantitative structure-activity relationship (QSAR) approach is most widely used for the prediction of biological activity of potential medicinal compounds. A QSAR model is developed by correlating the information obtained from chemical structures (numerical descriptors/ independent variables) with the experimental response values (the dependent variable). In the current study, we have developed a QSAR model to predict the inhibitory activity of small molecule carboxamides against severe acute respiratory syndrome coronavirus (SARS-- CoV) 3CLpro enzyme. Due to the structural similarity of this enzyme with SARS-CoV-2, the causative organism of the recent pandemic, the former may be used for the development of therapies against coronavirus disease 19 (COVID-19). The final multiple linear regression (MLR) model was based on four two-dimensional descriptors with definite physicochemical meaning. The model was strictly validated using different internal and external quality metrics. The model showed significant statistical quality in terms of determination coefficient (R2=0.748, adjusted R2 or R2 adj = 0.700), cross-validated leave-one-out Q2 (Q2=0.628) and external predicted variance R2 pred = 0.723. The final validated model was used for the prediction of external set compounds as well as to virtually design a new library of small molecules. We have also performed a docking analysis of the most active and least active compounds present in the dataset for comparative analysis and to explain the features obtained from the 2D-QSAR model. The derived model may be useful to predict the inhibitory activity of small molecules within the applicability domain of the model only based on the chemical structure information prior to their synthesis and testing.

Quality of life and its association with current substance use, medication non-adherence and clinical factors of people with schizophrenia in Southwest Ethiopia: a hospital-based cross-sectional study

BackgroundSchizophrenia was ranked as one of the top ten illnesses contributing to the global burden of disease. But little is known about the quality of life among people with schizophrenia, in particular in low-income countries. This study was aimed to examine the association of quality of life with current substance use, medication non-adherence and clinical factors of people with schizophrenia at Jimma University Medical Center, psychiatry clinic, Southwest Ethiopia.MethodsInstitution based cross-sectional study design was employed. Study participants were recruited using a systematic random sampling method and a sample fraction of two was used after the first person was identified by the lottery method. we used the World Health Organization Quality of Life Scale-Brief version (WHOQoL-BREF) and 4-item Morisky Medication Adherence Scale (MMAS-4) to assess the quality of life and medication non-adherence respectively. Data about current substance use was assessed by yes/no questions. Descriptive statistics, such as frequency, mean and standard deviations were computed to describe the characteristics of the study population. Data entry was done using EpiData version 3.1 then exported to SPSS statistics version 25 for analysis and analyzed using multiple linear regression. The assumption for linear regression analysis including the presence of a linear relationship between the outcome and predictor variable, the test of normality, collinearity statistics, auto-correlation and homoscedasticity were checked. Un-standardized Beta (β) coefficients with 95% confidence interval (CI) and P-value < 0.05 were computed to assess the level of association and statistical significance in the final multiple linear regression analysis.ResultIn this study 31.65% of participants were medication non-adherent and total mean scores of quality of life showed a lower level of satisfaction in social relationship domain (10.14 ± 3.12). Our study showed 152(43.3%), 248(70.7%) and 97(27.6%) of respondents had used tobacco, Khat and alcohol atleast once during the past 3 months respectively. Final adjusted multiple regression model showed medication non-adherence has significant negative association with physical domain (beta = − 4.42, p < 0.001), psychological (beta = − 4.49, p < 0.001), social relationships (beta = − 2.29, p < 0.001) and environmental domains (beta = − 4.95, p < 0.001). Treatment duration has significant negative association with psychological domain (beta = − 0.17, p < 0.04), social relationship (beta = − 0.14, p < 0.005), environmental domain (beta = − 0.24, p < 0.02) and overall quality of life (beta = − 0.67, p < 0.02). Having comorbid physical illness has significant negative association with physical domain (beta = − 2.74, p < 0.001), psychological (beta = − 2.13, p < 0.004), social relationships (beta = − 1.25, p < 0.007), environmental domain (beta = − 3.39, p < 0.001) and overall quality of life (beta = − 9.9, p < 0.001). Current tobacco use has significant negative association with physical domain (beta = − 1.16, p < 0.004), psychological (beta = − 1.23, p < 0.001), social relationships (beta = − 0.88, p < 0.001), environmental domains (beta = − 1.98, p < 0.001) and overall quality of life (beta = − 5.73, p < 0.001). Also, current chewing khat has significant negative association with physical domain (beta = − 1.15, p < 0.003), psychological (beta = − 1.58, p < 0.001), environmental domains (beta = − 2.63, p < 0.001) and overall quality of life (beta = − 6.22, p < 0.001).ConclusionThe social relationship domain of quality of life has the lowest mean score. Medication non-adherence, treatment duration, having a comorbid physical illness, current tobacco use and current chewing khat were found to have a statistically significant association with the overall quality of life. Therefore, treatments aimed to improve social deficits, medication non-adherence, comorbid physical illness and decrease substance abuse is imperative.

Predicting hydrolysis kinetics for multiple types of halogenated disinfection byproducts via QSAR models

In water and wastewater engineering systems, hydrolysis is one of the most important means in abiotic degradation of disinfection byproducts (DBPs). Enhanced knowledge of hydrolysis of DBPs can help determine the likelihood of DBP occurrence in water and their risks. In order to better understand the roles that functional groups play in the fate and occurrence of DBPs in the environment, this study developed several quantitative structure-activity relationship (QSAR) models to estimate the hydrolysis rate constants for a pool of 40 halogenated, small molecular weight DBPs originating from seven chemical classes. The models are based on the descriptors calculated and filtered by stepwise discriminant analysis with multiple-linear regression (MLR) and artificial neural network (ANN) algorithms, and overall they exhibited better performance than models using conventional descriptors only. The relative importance of selected descriptors are ranked as average mole weight > number of fluorine > polar effect of functional group > mole weight > number of chlorine > dipole moment > the ratio of oxygen to carbon atoms. The squared regression coefficients between predicted and experimental values were 0.939 and 0.976 for the final MLR and ANN models, respectively. While the ANN model demonstrated better performance than the MLR model for all 40 DBPs, the MLR model shows more accurate predictions for haloethanes, haloacetic acids, and haloacetamides. Unlike existing QSAR models, which treat each type of DBP separately, the models developed in this study are unique in considering multiple types of DBPs together; therefore, these models may help researchers to better understand the effects of not only halogens but also functional groups on DBP stability.

Structure-flammability relationship study of phosphoester dimers by MLR and PLS

Polyphosphonates and polyphosphates having good flame retardancy represent an important class of organophosphorus based polymer additives. In this analysis the flammability of 28 previously synthesized polyphosphoesters, modelled as dimmers, was explored using the multiple linear regression (MLR) and Partial Least Square (PLS) methodology. The statistical quality of the final MLR and PLS models was estimated using the following parameters: the squared correlation coefficient ( r t r a i n i n g 2= 0.917 and 0.976), the training root-mean-square errors (RMSEtr = 0.029 and 0.016) and the leave-seven-out cross-validation correlation coefficient ( q L 7 O 2= 0.748 and 0.881), respectively. External validation was checked for a test set of seven compounds using several criteria. The MLR models had somewhat inferior fitting results. The final MLR and PLS models can be used for the estimation of limiting oxygen index (LOI) values of new polyphosphoester structures. The presence of phosphonate groups and increasing molecular branching in an isomeric series favour the dimer flammability.

Factors associated with differences in perceived health among German long-term unemployed

BackgroundUnemployment is associated with reduced physical and psychological well-being. Perceived health is an important factor influencing health outcomes as well as successful returns to work. This study aims to determine the extent to which perceived health correlates with mental health, various health risk characteristics and socio-demographic characteristics in a setting-selected sample of long-term unemployed persons.MethodsUsing SF-12, 365 long-term unemployed persons were assessed for self-perceived health and various socio-demographic and health characteristics. Perceived health data of the sample was compared to the German SF-12 reference population. Bivariate analyses and multiple linear regression models were applied to identify those variables significantly associated with perceived health.ResultsThe study population reported poorer perceived health compared with the general population. Analyses showed that perceived mental health was significantly worse in women, among persons with heightened depression and anxiety scores, and in participants reporting reduced levels of physical activity. Perceived physical health was significantly lower among older persons, participants with a higher BMI, and participants with heightened depression and anxiety scores. Both mental and physical health were worse among the unemployed assigned to an employment center as compared to those engaged in the secondary labor market. In total, 36% of the variance in the SF-12 mental score and 20% of the variance in the SF-12 physical score were explained by the factors included in the final multiple linear regression models.ConclusionsPerceived health among a select group of long-term unemployed is reduced to a clinically relevant extent compared to the general population. The preliminary findings underline an association between mental health and perceived health. Negative self-perceptions of health were also associated with the labor market setting and some of the socio-demographic and health behavior variables. Further research is needed to determine risk factors leading to reduced perceived health in the unemployed. The strong association between mental health and perceived health suggests interventions targeting mental health are urgently needed to positively influence perceived health, a key determinant of individuals’ chances to successfully return to work.

On the response of Halogen Occultation Experiment (HALOE) stratospheric ozone and temperature to the 11‐year solar cycle forcing

Results are presented on responses in 14‐year (1991–2005) time series of stratospheric ozone and temperature from the Halogen Occultation Experiment (HALOE) of the Upper Atmosphere Research Satellite (UARS) to a solar cycle (SC‐like) variation. The ozone time series are for ten, 20‐degree‐wide, latitude bins from 45°S to 45°N and for thirteen “half‐Umkehr” layers of about 2.5 km thickness and extending from 63 hPa to 0.7 hPa. The temperature time series analyses were restricted to pressure levels in the range of 2 hPa to 0.7 hPa. Multiple linear regression (MLR) techniques were applied to each of the 130 time series of bin‐averaged, sunrise plus sunset ozone points over that latitude/pressure domain. A simple, 11‐year periodic term and a linear trend term were added to the final MLR models after their seasonal and interannual terms had been determined. Where the amplitudes of the 11‐year terms are significant, they are in phase with those of the more standard proxies for the solar UV flux. The maximum minus minimum response for ozone is on the order of 2 to 3% from about 2 to 5 hPa and for the latitudes of 45°S to 45°N. There is also a significant maximum minus minimum response on the order of 1 K for temperature between 15°S and 15°N and from 2 to 0.7 hPa. The associated linear trends for ozone are near zero in the upper stratosphere. It is noted, however, that effects of the changes in total chlorine during this 14‐year period were not accounted for explicitly and that their omission can be a confounding effect for both the analyzed solar cycle responses and the linear terms, especially for tropical upper stratospheric ozone. It is concluded that the solar occultation technique of HALOE provided adequate sampling and sufficient vertical resolution for obtaining the solar cycle response in stratospheric ozone.

A reanalysis for the seasonal and longer‐period cycles and the trends in middle‐atmosphere temperature from the Halogen Occultation Experiment

Remsberg and Deaver (2005) reported on multiple linear regression (MLR) analyses of time series of middle‐atmosphere temperature versus pressure (or T(p)) profiles from the Halogen Occultation Experiment (HALOE) for its seasonal and longer‐period cycles. Their results are extended herein to just over 14 years and updated to properly account for the effects of autocorrelation in their time series of zonally averaged data. The updated seasonal and annual average terms are provided, and they can be used to generate near‐global, temperature distributions from 2 to 0.007 hPa that are representative of the period 1991–2005. Quasi‐biennial oscillation (QBO‐like; 853‐day) and subbiennial (640‐day) terms are also resolved and provided, and they exhibit good consistency across the range of latitudes and pressure altitudes. Further, somewhat exploratory analyses of the residuals from each of the 208 time series yield significant 11‐year solar cycle (or SC‐like) and linear trend terms at a number of latitudes and levels. Where significant, those terms are included in the final MLR models. The amplitudes of the SC‐like terms for the mesosphere agree reasonably with calculations of the direct solar radiative effects for T(p). Those SC‐like amplitudes increase by about a factor of 2 from the lower to the upper mesosphere and are larger at the middle than at the low latitudes. The diagnosed T(p) cooling trends from HALOE for the low latitudes are in the range of −0.5 to −1.0 K/decade, which is in good agreement with the findings from models of the radiative effects on pressure surfaces due to known increases in atmospheric CO2. The diagnosed HALOE trends are somewhat larger than those predicted with models for middle latitudes of the upper mesosphere. Cooling trends also are found to be increasing from about −0.5 to −1.0 K/decade from 1 to 2 hPa of the upper stratosphere, also in reasonable agreement with modeled results.

EFFECT OF MORBIDITIES ON COSTS OF END‐STAGE RENAL DISEASES (ESRD) PATIENTS IN THE TOP END OF THE NT

Objectives The continuing care cost of maintaining ESRD patients includes not only the cost of dialysis, but also the costs of treating comorbidities related to dialysis treatments (MRD) and not related to dialysis treatments. This study examined the relationship between comorbidities and costs. Methods: Cost information of ESRD patients was derived from Royal Darwin Hospital costing study in 1996/97 and 1997/98 financial years. Diagnosis Related Groups (DRGs) assigned to patients who were hospitalised were re‐categorised as 19 groups of comorbidites. Multiple linear regression (MLR) model was developed to analyse the relationship between diseases and costs. A forward selection strategy was applied to select diseases significantly affecting costs. Results: In that 2‐year period, 167 people (102 Aboriginals (A), and 65 non Aboriginals) were treated for ESRD, at a total cost of $12.7million. Provision of dialysis treatment alone cost $9.7m (76% of total cost). One hundred and thirteen people (85% of A, 40% of non A) had intercurrent hospitalisations for comorbidities at a cost of $3m (24%), with 530 episodes. Of these episodes, the most frequent reasons of admission for A were: water and electrolyte disorder (WED 25%), respiratory diseases (RD 20%), and surgical procedure (13%), while for non A, they were cardiovascular diseases (CVD 27%), WED (17%) and SP (15%) (see below). The following diseases (p &lt; 0.05) were chosen in the final MLR model in terms of cost: SP, RD, CVD, WED, MRD, and access procedure etc. SP, RD and CVD accounted for 30%, 18%, and 11% of the $3m. % of episodes SP RD CVD WED MRD Access AB (% of 422) 13 20 12 25 5 0.9 non AB (% of 88) 15 7 27 17 5 5 Cost % 30 18 11 9 6 5 ConclusionsTo reduce the continuing care costs for ESRD patients, greater efforts must be made to decrease the comorbidities. Prevention of surgical diseases, respiratory infection, and heart diseases might reduce the morbidity rate in ESRD patients, and subsequently reduce costs.