Scribd Logo
Upload

Welcome to Scribd!

  • Untitled

    Uploaded by

    Vidya Ikawati
    22 views19 pages

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    Download as pdf
    22 views19 pages

    Untitled

    Uploaded by

    Vidya Ikawati

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    Download as pdf
    of 19
    ‘www.nature.com/scientificreports scientific reports OPEN (Por or on Development and validation of a hypertension risk prediction model and construction of a risk score in a Canadian population Mohammad Ziaul Islam Chowdhury”, Alexander A. Leung", Khokan C. Sikdar?, Maeve O’Beirne?, Hude Quan® & Tanvir . Turin!” Identifying high-risk individuals for targeted intervention may prevent or delay hypertension onset. ‘We developed a hypertension risk prediction model and subsequent risk sore among the Canadian population using measures readily available ina primary care setting. A Canadian cohort of 18,322 participants aged 35-69 years without hypertension at baseline was followed for hypertension incidence, and 625 new hypertension cases were reported. At a2:1 ratio, the sample was randomly divided into derivation and validation sets. In the derivation sample, a Cox proportional hazard model was used to develop the model, and the model's performance was evaluated inthe validation ‘sample. Finaly, a risk score table was created incorporating regression coefficients from the model. ‘The multivariable Cox model identified age, body mass index, systolic blood pressure, diabetes, total physical activity time, and cardiovascular disease as significant risk factors (p<0.05) of hypertension. incidence, The variable sex was forced to enter the final model. Some interaction terms were identified ‘a significant but were excluded due to their lack of incremental predictive capacity. Our model showed good discrimination (Harrel'sC-statistic 0.77) and calibration (Grennesby and Borgan test, 1 statistic= 8.75, p=0.07; calibration slope 1.006). point-based score forthe risks of develo hypertension was presented after?2,,3-, 5-, and 6 years of observation. This simple, practical prediction score can reliably identify Canadian adults at high risk of developing incident hypertension inthe primary care setting and facilitate discussions on modifying this rzk most effectively. Hypertension, which affects more than 1 in 5 Canadians', is a common medical condition and isthe leading modifiable risk factor for preventable cardiovascular morbidity and mortality” Hypertension prevention and blood pressure management in hypertensive patients i considered major public health concern For dec the focus of interventions has been on improving hypertension detection, treatment, and control but relativ litle work has been done to promote primsry’ prevention, Evidence suggests that the risk of progression to hypertension depends on several factors. Older age, female sex, increased body mass index (BMI), family his tory of hypertension, premature cardiovascular disease, sedentary ifestyles, unhealthy diet, and high sou ‘consumption are among the factors reported as significant predictors of hypertension Serecning people at greater risk of hypertension opens the possibilty of promoting individualized preventive initiatives because we wll know who to target, what to target, where to target, and how totarget™. A prediction model helps sereen high-risk individuals by estimating their probability of developing hypertension within a particular time’ Over the past decades, many prediction models have been developed in different populations to predict incident hypertension”, but their performance in accurately forecasting it varies. To the bes of our knowledge, prediction models forthe risk of incident hypertension that directly address the Canadian popula tion have not yet been established. One method for predicting the risk of developing hypertension in Canadian "Department of Community Health Sciences, University of Calgary, 3280 Hospital Drive NW, Calgary, AB T2N, 426, Canada. ‘Department of Family Medicine, GOI2F, Health Sciences Centre, Cumming Schoo! of Medicine, University of Calgary, 3330 Hospital Orive NW, Calgary, AB T2N 4Ni, Canada. “Department of Peychiatry, University of Calgary, 3280 Hospital Drive NW, Calgary, AB T2N 426, Canada. Department of Mecicine, University ‘of Calgary, 3260 Hospital Drive NW, Calgary, AB T2N 4Z6, Canada, "Health Status Assessment, Survllance and Reporting, Pubic Heath Survelance and infrastructure, Provincial Population and Public Health Alberta Health Services, 10101 Southport RE. SW, Calgary, AB T2W 3N2, Canada, “emall:chowahut@ucalgary ca SclentifeReports|_ (2023)323780 [itp oi rg/102038)+41599-022-16904-x natureportoo| www.nature.com/scientificreports/ Scientific Reports | populations was to usean existing model and evaluate is performance through external validation ofthe model Hovrever, in light ofthe following consklerations, we opted to construct a new model rather than externally validate an existing model, First, prediction models are determined by an equation that includes tsk factors, risk coefficients (multiplying factors that assign an etiological weight to single factors), and the general popula tions survival probability or baseline risk without the disease". These elements vary depending on the type of population, especially whea very different cultures are compared (.e, European countries and Asian counties). Second, each population has a different risk of contracting the disease, and each population may havea diferent dlistribution of rsk factors that weigh differently in determining the disease". Furthermore, the disease may ‘occur with varying probability, resulting ina diferent survival rate without i, Third, heterogeneity in predictor ‘elects (the same predictor may have different prognostic values in different populations), differences in outcome incidence, andl differences in case-mix between the development and validation cobort can all have a significant impact on a models predictive performance and frequently result in poor performance when applied toa dif. cent population’, Furthermore, many existing models were restricted to people ofa specific ethnicity or those ‘who were already at high risk or only inluded a limited numberof clinical variables. Because ofthese facts, the Performance of a prediction model ean vary significantly by population, Asa result the prediction model's acc acy is frequently acceptable for that index population but is not necessarily generalizable to populations other than the one for which the model was developed, We assessed this by applying a few published hypertension prediction models to our population and comparing their predictive performance. Prediction models cannot be transferred directly om one population type to anther” =, The lack ofa hypertension risk index specific tothe ‘Canadian population prompted us to create anew hypertension predietion model using one of Canada’ largest cohort studies, which will aid local clinicians and healtheare providers in clinical decision-making, planning, and proper management of hypertension-related healthcare services. Faced with th lowest national rates of blood pressure contol in over a decade, elective strategies to identify ‘Canadians at the highest risk of developing high blood pressure to prevent the onset of hypertension have become ‘more relevant than ever. To this end, we erated and internally validated a simple and practical risk prediction ‘model for incident hypertension in the Canadian adult population. We also derived the pointbase risk score from the developed model to facilitate clinical practice use for decision-making. Methods Study population. ‘The study subjects were from Alberts Tomorrave Project (ATP) cohort data, ATP isa province-wide prospective cohort study and consists of Alberta residents, aged 35-69 years, without any his tory of cancer, other than non-melanoma skin cancer, ATP contains baseline and longitudinal information ‘on socio-demographic characteristics, personal and family history ofthe disease, medication use, lifestyle and health behavior, environmental exposures, and physical measures, ATP was designed to be representative of| healthy midale-aged adults in Alberta. A more detailed description of ATP and its recruitment process is pro vided inthe supplementary material (Appendi 1). ‘Our study cohort consists of 25,359 participants who completed ATP CORE questionnaire and consented to have their data linked with Allertals administrative heath data Linking with administrative health data was «done to establish the necesary longitudinal follow-up to determine hypertension incidence. We excluded 6996 ‘participants from the analysis who had hypertension at baseline and di not meet eligibility criteria (Le, ree of hypertension at aseline). We also excluded 41 partieipants who responded to hypertension status questions at [baseline as “don't know” or missing’ Eighteen thousand three hundred twenty-two participants were inchided inthe final analysis, Selection of candidate variables. Before commencing the analysis, we compiled alist of available poten tial candidate variables. We determine the possible candidate variables for inclision in model development ‘based on aliterature search, variables that have been used in the past, and discussion with content experts. For this std, we considered 29 candidate variables for inclusion in the model, We deliberately didnot consider ‘genetic sk fctorsbiomarkers as potential candidate variables given our model’ intended clinical aplication Inclusion ofthe genetic risk factors inthe model ca reduce the models usability due to a lack of readily avail: able information Definition of outcome and variables. The outcome incident hypertension was determined from linked administativehealth data using a coding algorithm. We used the relevant ICD-9 and ICD- 10 codes (ICD-9-CM codes: 401.x, 402.x, 403.x, 404x, and 405.x; ICD-10-CA/CCI codes: I10.x,ILL.x, 12, 113x, and 115.x) and validated hypertension case definition (two physician claims within two years or one hospital discharge for hypertension) to define hypertension incidence (sensitivity 75%, positive predictive value 81%)" ‘Out of29 candidate variables, 1] were continuous, and 18 were categorical. Continuous variables remained ‘continuous in the model developed and categorized only fr deriving risk scores, A detailed description of the ‘variables and their categorization is provided in the supplementary material (Appendix 2). values. Qurdataset has missing values on several candidate variables ranging from 0 to 26%, Infor. ‘mation on missing values for different candidate variables is presented in the supplementary table Table SI). We ‘sed multiple imputation for missing data". Ths technique predicts the missing values by ubiiting the existing information from other available variables and then substitute the missing values with the predicted values to create a complete dataset. Multiple imputation by chained equations (MICE) was used to impute the missing values using Stata "ice command™. (aoz2)32:2780 | ep ot or/l0 1036)41598-022-16508-« natureportotio www.nature.com/scientificreports/ Scientific Reports | Statistical analysis. Before imputing missing values, the required assumption “missing at random” for per foraning multiple tnpatations was checked. We compared the study characteristics of those with missing with those without mising information using appropriate tess (unpaired (testo the x test). Cominuous variables were expressed a the mean (SE) and alegrical variables were expressed as numbers (percentage ofthe ttl. ‘We randomly split subjects into to sets: the derivation set, which included 67% (two-thirds) ofthe sample (12.23), an the validation st, which iced the remaining 33% (one-thid) (a™~689).The two groups! ‘baseline characteristics were compared vsing the unpaired (esto theta, a appropriate. We developed 4 vs prediction model from the derision dats using the mmltvariale Cox proportional hazards model and assessed the goodnes offi using the validation data. Callinerity among the variables was tested using the variance inflaton factor (VIF) witha threshold of 25% From the list of candidate variables, highly correlated variables were exchided based on VIF before applying the model “The univariate Cox proportional hazards model was appli fist to screen the variables fora signiicant association {p< 0.20)" with hypertension incidence inthe devvation set. Variable identifed as significant the “ivan astociation were later pt into a multivariable Cox proportional hazards model to determine timate Sgntcant risk actor (p03) of incident hypertension. The interaction ters were lo tested with significant ‘arabes identied in the multivariable Cox mall. Daring the model development proces, the proportional hazard assumption associated with the Cox model was also tested. There are several methode for verifying pro Portonality assumption, and we tested the proportionality assumption by using the Schoenfeld and scaled Sch ‘enfeld residuals. We ested the proportionality of the model as a whole and proportionality foreach predictor. "The following general equation was used to calculate the rik of incident hypertension within Ge Probability where St) isthe baseline survival function, assuming all variables are represented by average values at follow. up timer; fis the estimated regression coeficient ofthe ith variable; X;is the value ofthe it variable; Xyisthe ‘corresponding mean, and p denotes the numberof variables, In the validation data, the model’ predictive performance was assessed. Model discrimination was evalu: ated using Horrells C-statistic™. Harrels C-statisti indicates the proportion of all pirsof subjects that can be ‘ordered such that the subject who survived longer will have the higher predicted survival time than the subjects who survived shorter, assuming that these subject pairs are selected at random, Calibration was asscsted Using the Gronnesby and Borgan (GB) test”. The GB test isan overall goodness-of-fit test fr the Gox proportional hazards model and is based on martingale residuals. In the GB tet, the observations are divided into K groups according to their estimated rsk sore, an approach similar to Hosmer and Lemeshove goodness-of fit for logistic Fegression™ Bricr score was calculated st diflerent time points, and calibration plot was also used for asses ing calibration. Ina calibration plo, expected probabilities (predicted prababiltes from the model ae plot ted against observed outcome probabilities (calculated by Kaplan-Meier estimates). Arja like plots were used for assessing the goodness offit graphically”. We also produced histograms ofthe prognostic index (a linear predictor of the Cox model) to show the prognostic index distribution in the derivation and validation data {et We also assessed calibration using the approsch proposed by Royston P™, where observed (Kaplan-Meier) and predicted survival probabilities compared in some prognostic groups derived by placing cut points on the prognostic index. We defined three risk groups (good, intermediate, and poor) from the 25th and 75th eeniles ‘ofthe prognostic index in the derivation dataset based on events, ‘We then created a point-based scoring system from the model so that itcan be easly used in clinical practice Integer points were assigned according tothe presence/absence ofeach rsk factor so that the overall risk can ‘be estimated by summing the points together. We constructed the risk score utilizing the regresion coeficients ‘of our Cox model according to the method proposed by Sullivan etal”. To facilitate calculating risk sco ‘continuous variables considered in the model development were divided into categories as discussed befor. All statistical tests were two-sided. All statistical analyses were performed using Stata (Version 15.1 Corporation, College Station, Texas 77845, USA), ‘Comparing existing model performances to the developed model. We used a few existing hyper: tension risk models in our datast to explain how our developed mal performed in comparison to those mod! cls when those were applied to our poptlation. Model selection was primarily made based on the availability of the final variables considered in those selected models in our dataset. This eliminated the majority of existing ‘models from consideration for validation in our dataset. In addition, whether the model provided enough information to perform the validation was also considered a major factor in selecting a model For example, ‘A model did not provide regression coefficients or hazard or odds ratios from which coefiients can be deri ‘were excluded from consideration. Also ifa model didnot provide te predictive perlormance ol thei models, such as discrimination or calibration they were exchuded from considerations. In this case, we wont be able 10 ‘compare the vlidated models predictive performance in our dataset, Considering the aforementioned factors and information from our recent systematic review", we selected the models by Parikh etl”, Kivimiki etal”, Lim etal." Chien etal", and Wanget "for validation in our dataset, The model by Parikh etal. also known asthe Framingham Risk Score (FRS), was developed inthe United Sates in a predominantly White population, ‘ith age, sex, systolic blood pressure (SBP), diastolic lood pressure (DBP), BMI, parental hypertension, Fette smoking, and age by DBP as final variables. Kivimaki etal.” developed the Whitshall It Risk Score in Eng landin predominantly White population, In model construction, the same FRS variables were used, Lim etal developed their model in Korea in an Asian population using the same variables as FRS. Chien et al.!“developed (aoz2)32:2780 | ep ot or/l0 1036)41598-022-16508-« natureportotio www.nature.com/scientificreports/ Scientific Reports | thee model in Taiwan among the ethnic Chinese population, Two models were created, and we validated their clinical model using age, gender, BMI, SBP, and DBP asthe final variables. Wang etl developed theie model in China with a rural Chinese population, The final variables in the model were age, pasental hypertension, SBP, DAP, BMI, and age by BML. The final variables considered in these models were available in our dataset. “This study’ ethies was approved by the Conjoint Health Research Ethics Board (CHHREB) at the University of (Calgary, and all methods were performed in accordance with the relevant guidelines and regulations. Informed, consent was waived by the CHREB (REBI8-0162_REN2) because the dataset used in this study consisted of dde-denifed secondary data released for research purposes Patient consent. Not required, The manuscript is based on the analysis of secondary de-identified data, Patients and the public were not involved inthe development, design, conductor reporting of the study. Results Baseline characteristics ofthe sty participants ae presented in Table | and supplementary table (Table $2). In Table 1, the study participants characteristic are given forthe entre cohort aswell ax compared according, to the status of developing hypertension. In contrat in Table S, characteristics are compared between the ‘derivation sample and the validation sample. Overall the study participants mean age wss 5099 years and the puticipation of females (68.55%) in the study was higher than the males (31 45%). During the median 5. year follow-up, 625 (3.41%) participants developed hypertension. In Table 1, most ofthe study characteristics were significantly ciferent (p.=0.05 between those who developed hypertension and those who did not. Those who Aleveloped hypertension were relatively aler, had higher (average) BMI, DBP. SBE, and more with diabetes and ‘ardiasculareisease. the proportions of males and females were also significantly diferent between these two {roups, However some sly characteristics were similar with no statistically significant diflerence (p> 005), inching ethnic amily history of hypertension, alcohol consumption, and total physical activity time. When ‘we randomly divided the date nto derivation and validation sts Table S2), the stad characteristic ere smile vith no significant ference (p 0.05) between the deevaton and validation sample except BMI waist ati. From thelist of candidate variables, six (ever smoked, hip circumference, bodyfat percentage, BMI waist ratio, wast cizcunfeence diastolic blood presure) were excluded from the model bung due to tei high collinearity (threshold VIF> 25) with other variables. Campaing the study characteristics between the missing and imputed is presented in the supplementary table (Tables), Inthe derivation sample, most ofthe candidate varsbles used in our study were identified a significant univariate predictors (Table 2) Variables not sgniicantly associated with incident hypertension in uniariate models were excluded from the multivariable model. In the multivariable modal, ae, sex, BMI, SBP. diabetes, CVD, total physical activity time, depression, waist-hip rato residence, highest education lve completed, work ing status, otal household income, family history of hypertension, smoking statu, total sleep ime, vegetable and {rit consumption, an job schedule was included The multivariable Cox model indicated that age, BMI, SHR, dabets, ttl physica activity time, and cardiovascular disease wee independent sk factors of nck ion Table 2}, We forced sex into the model, considering it incl importance. The follwing itera were aed othe model with other significant variables nthe multivariable Cox model: age by BMA age by SBP, age by diabetes, age by CVD, age by total physieal activity time, ge by sex, BMI by se, SBP by ex, “labs by sex, CVD by sex and total physical activity time by sex. When the interaction terme were inckuded inthe model, age by sex, age by BMI, age by SBP ae by total physical activity time, sex by SBP, and sex by CVD showed significant association wih incident hypertension (Table3) However, the inclusion of these interaction terms did not improve the models’ discriminative performance. The models with and without interaction ere were vctally identical regarding their Harrels statistics value (0.77 and 0.77, respectively) and statistical Significance (p=0.6). Consequenty the interaction terms wer excluded irom the nll selected model. The ‘model with only main effects was used in subsequent analyses to construct a simpler and more user-friendly Fisk estimation equation and risk score. A global est for Cox proportional hazas assumption indicated no Violation af assumptions (p=0.72) (Supplementary Tae St). The bain survive function at median llow-up lime 3. 80-years = 6-year (6) was (0.97) Inthe derivation sample, the models discriminative performance (HarresC-statistic) was 077 ‘When we aplied our derived model in the validation sample, the models discriminative performance was good (Harte statistic 0.77). The results ofthe GB test indeatedan aceptablecallation ofthe risk pred tion model (statistic 875, p=0.07, Fig. 1). To compare the observed and expected events in each group based onrak score, Ara ike plots are also presented (Fig 2) A calibration plo four prediction model at tie of {rycors was also presented in Fig 3. A calibration slope of 1.006 indeates that predicted probabilities donot vary enough Figuee 4 repeesnts the calibration of our model in the derivation and validation datasets. The ‘alibration of the model looks good in cach dataset. The predictions inthe validation dataset are good for both “Good” andIntermediat” isk groups where survival and predicted probabilities te quite simia, xcept ightiy higher predictions between 6- and 11-years ime intervals for theItermecate’ group. The predictions in the “Poor” group are consistent withthe survival up to year stand somewhat high ate: thats, srvial tends toe ‘worse than predicted. Duco fewer validation data events the confidence interval tend te wider in validation data than i the derivation data Figure S presents the prognostic index histogram in derivation and validation ‘ata and no obvious regularities and outers were detected, Brier coreclculatd at 4-year, 5-year 6 yea, ad 7-year ime points ate 0.018, 0021, 0.026, and 0.029, respectively indicating accurate predictions. Final fom the developed model, a imple ad practical risk score was created to calelate the risk nc diferent times (2-year 3-yeae 5-year, and 6-year) (Table 1). The constant forthe points her of regression units that wll corzespond to one pont wa seta thers sesocited witha (aoz2)32:2780 | ep ot or/l0 1036)41598-022-16508-« natureportolio ayn) 99 (007) 91035 (007) oo lara a e) ‘Wr GE) 548 0079) {Sinan ims 0877 2156408 Ss (76 aoot leateeinis) Stolesthan 5190 28.33) 226 (36.16) 4964 (28.05), 368 a9 a6 09) ae Nal we0K 25000) S015 aay - Female 12,559 (68.55) 375 (60) 12.184 (68.85), ey de enn 2615000 2869 020 2638000 aon ade 8) mom z 5 rm GRE-2099) THB Cay woe Fa) ming.) | oon Body mas inderKeIm2+ 0%) | Creighl 250-2095) | 68765755) 8.7) ‘501 0673) O90) Sui) mes) SC) wa in an azarae) ain (003 oan 000) aon aie aso) aasiome 025 oom oa wastaininqurtie, | urte? ezrome) 227 00069 027 0a) oa non) aes emo) a> rome) oo on Quiet e200) asvcomie) 00mm co Tipcintce aa To4as (0a Tos 0a) Too @09) nr ‘i amir, aa 070) Tee 0) 221 00) ), we used the Ist percentile and the 99th percentile ofthat variable to minimize the influence of extreme values, The points were initially computed asa decimal value, but later rounded to the nearest integer for facile calculation. The approximate risk of incident hypertension was then estimated via summation of the points awarded to each of the items, We attach the risks associated wit ‘each point total using the Cox regression equation (Table 5). Finally, we created risk categories according to the total points In our model, the maximum total points 4, and the minimum is~2. For simple interpretation in clinical setting, we categorize estimated risk into three categories and presented in Table 6, Case study. 50-year-old male with BMI 285, SBP 135, diabetic, no CVD, and moderate physical activity (850 MET minutes/week). ee 2 Sox ° a 3 Die ae ee 1 ys act Moderate 020 MET minted) "he tina oyu) 731 “The isk timate based on our newly developed Cox model is computed as follows: Six; = 0.02768(50) + 0.08722(0) + 0.05147(28.5) + 0.04629(135) + 0.57066(1) + 1.08710(0) ~ 0.00003(850) = 9.645205 Yo Ax, = o02768 5094) + 008722003142) + 005147(2648) + 0.04629119.75) “0576100 + Lom 0 02) —oBNERAISIT) = RII so pS) pai-s” ScientifeReports| (2022)3222786 | ep ot or/l0 1036)41598-022-16508-« natureportotio —— TT oo agora) oo ia ve ro em ea a5 i tor 1 co am Tare 8 Sar Tost 1 ar ae topo - asa | om Tipo rasa 08 Wa crf iat) Em Waste at ona 8) tom 5H Fa sm Fo ra ae rasta Er Distal ed ese me tows.) aor Syd pee i iss) Sat ra ae sania ror i ‘ao ara aT comme Cie wi pal ist) oy Te fl ru ar Poors) Co Taal roy 1 ao a eas 8) 0 gy, OORT Com Siox00-519.989 asi 038.065) oan |<" Farscasr-039) Com =a ear a ar ase aa | et cn oid Tigh colar ome deny oa WD (eta son ee aa ame — ingen cushy ChE nay 79 063.09 aso | <2 | or aze.ay eon oa CPCS cs Ce manne ener reneraietederme Tosa a4s-o69) “<0 082 06-100) ons tenekn nna iy abet ema oo ae ot a a i ot Fp Waray aed aa Tipe eas 5 Co tet ear oat Cie nian aa rat a5) er ta % oo ‘as = Ein em Ta i Cart co Sa Er 2a AY ar Deion co oo ro ear OT TS oa Fa iperesn = eas aes te LH OI) ee LO ae ig ae ee teen me waar Joon ore) aaron = ry Tees ry oan SciniteReports|_ (033)25780 | ‘ips ior 108 4158-02 6804 tani porto ‘www.nature.com/scientificreports/ fs 19415 09 Akai oompon tine e400 ums Bettina a6 059-130 aie 020 ‘estine sek aca) 0 Sense oan0-100 jase Fate Fe ee rte a aa) as (0-1) a i ig ero, a et INL T3329 Si——| tafe =i va ime —[ Diabetes ee 7 cnc | orn PE —-— Ua CRAMER as) Ye : Ey " Physica activity ttt | -o.00003, | Moderate (450-900. MET min- | eng 0.0120 1 ci pay ape EF |g saw = ‘able 4. Calculation of point values for risk score ‘Reference category. The age range inthe sample is 35-70. “The range of body mass index is 125-649, To determine the reference values for the stand las categories, wwe use the Ist percentile (183) and the 99th percentile (12.7) to minimize extreme values influence. “The ‘ange of physical activity total i fromm 83 MET sinuts/weck to 19.278 MET manutes/weck, To determine the reference values fo the frst and lst categories, we use the Ist percentile (9) and the 9th percentile (13,518) to minimize extreme values influence. "The range of systolic blood pressures is 76-205. To determine the reference values fo the fst and lst categories, we se the Ist percentile (02) andthe 9th percentile (156) to minimize extreme values influence The constant forthe point sytem or the numberof regression units willcorrexpond to one pont. Here we lt Befect the inereas in sk associated with a 5yeae increase in age: (0.02768) = 0.1384 ‘complete cases, excluding those with missing values, we imputed missing values in our study. This approach prevented information loss, maximized information utilization, and made the results robust ‘We could have used an existing model and evaluated its performance through external validation in our dataset before creating a new risk score. However, we refrain from doing ths forthe following reasons: First, prediction mode typically performs worse than it did with its original study tive accuracy is discovered alter an external validation study, researchers mist decide whether to reject the model oF update it to improve its predictive accuracy. By combining information ‘captured inthe original model with information from new individuals from the validation study, the model can De updated or recalibrate for local circumstances". Model updating entails adding more predictors or altering ne «portion of the formula to better suit the external population, The appropriateness of model updating dh «ternal validation. point of contention song researches, Some lim tha he researches ae develo ‘new prediction model even with minor changes "=", Second, developing a new prediction model along W ‘externally validating a well-known existing prediction model inthe development cohort and concluding that ‘the new model performs better isan inappropriate comparison in our view, Because this is then comparing the performance of one model in development to the performance of another model in external validation”. The evry developed model wll almost always appear superior because iti optimally designed to ft the develop ‘ment data The performance of two existing prediction models should be directly compared in an external validation dataset that is independent of both model development cohorts. Given this, we did not evaluate a ‘existing models performance and then develop a new model on the same dataset. Nevertheless forthe purpose ‘of comparison, we assessed a few ofthe published hypertension risk prediction models in ou population and. found that their performance was inferior to ours, ‘Our study has several limitations, Study participants were middle-aged Canadians. Prevention strategies are likely to be more effective ifthe young population can be targeted. Nevertheless, our study participants age range wil likely have minimal impact on our study's generalizability, as essential hypertension develops in the middle aged adults" as represented here. At baseline we excluded participants with self-reported hypertension, ‘which can potentially lead to misclassification of hypertension status, The incidence rate of hypertension in our Scientific Reports | (aoz2)32:2780 | ep ot or/l0 1036)41598-022-16508-« natureportolio oo ost 2a a so 7 a a7 -an10 aaa 038 28 Wet lege seals els elelelels elelslels 85 sl ele 09 78 am e@|sle]s als/e|s|s]a/e/e/e a ‘Tables. Risk estimates for point totals a 2, 3,5, and 6-year time. We determine the rsks that are associated ‘with each point in total The first step isto select the point totale’ theoretical range based on the point system ‘computed earlier. In our point system, the theoretical range of point totals s~ 2 to 40, We then attached a risk ‘’stimate to each point total using the Cox regression equation. -

    You might also like