E-Science technologies in a workflow for personalized medicine using cancer screening as a case study.

OBJECTIVE: We provide an e-Science perspective on the workflow from risk factor discovery and classification of disease to evaluation of personalized intervention programs. As case studies, we use personalized prostate and breast cancer screenings. MATERIALS AND METHODS: We describe an e-Science initiative in Sweden, e-Science for Cancer Prevention and Control (eCPC), which supports biomarker discovery and offers decision support for personalized intervention strategies. The generic eCPC contribution is a workflow with 4 nodes applied iteratively, and the concept of e-Science signifies systematic use of tools from the mathematical, statistical, data, and computer sciences. RESULTS: The eCPC workflow is illustrated through 2 case studies. For prostate cancer, an in-house personalized screening tool, the Stockholm-3 model (S3M), is presented as an alternative to prostate-specific antigen testing alone. S3M is evaluated in a trial setting and plans for rollout in the population are discussed. For breast cancer, new biomarkers based on breast density and molecular profiles are developed and the US multicenter Women Informed to Screen Depending on Measures (WISDOM) trial is referred to for evaluation. While current eCPC data management uses a traditional data warehouse model, we discuss eCPC-developed features of a coherent data integration platform. DISCUSSION AND CONCLUSION: E-Science tools are a key part of an evidence-based process for personalized medicine. This paper provides a structured workflow from data and models to evaluation of new personalized intervention strategies. The importance of multidisciplinary collaboration is emphasized. Importantly, the generic concepts of the suggested eCPC workflow are transferrable to other disease domains, although each disease will require tailored solutions.

Molecular Differences between Screen-Detected and Interval Breast Cancers Are Largely Explained by PAM50 Subtypes.

Purpose: Interval breast cancer is of clinical interest, as it exhibits an aggressive phenotype and evades detection by screening mammography. A comprehensive picture of somatic changes that drive tumors to become symptomatic in the screening interval can improve understanding of the biology underlying these aggressive tumors.Experimental Design: Initiated in April 2013, Clinical Sequencing of Cancer in Sweden (Clinseq) is a scientific and clinical platform for the genomic profiling of cancer. The breast cancer pilot study consisted of women diagnosed with breast cancer between 2001 and 2012 in the Stockholm/Gotland regions. A subset of 307 breast tumors was successfully sequenced, of which 113 were screen-detected and 60 were interval cancers. We applied targeted deep sequencing of cancer-related genes; low-pass, whole-genome sequencing; and RNA sequencing technology to characterize somatic differences in the genomic and transcriptomic architecture by interval cancer status. Mammographic density and PAM50 molecular subtypes were considered.Results: In the univariate analyses, TP53, PPP1R3A, and KMT2B were significantly more frequently mutated in interval cancers than in screen-detected cancers. Acquired somatic copy number aberrations with a frequency difference of at least 15% between the two groups included gains in 17q23-q25.3 and losses in 16q24.2. Gene expression analysis identified 447 significantly differentially expressed genes, of which 120 were replicated in an independent microarray dataset. After adjusting for PAM50, most differences were no longer significant.Conclusions: Molecular differences by interval cancer status were observed, but they were largely explained by PAM50 subtypes. This work offers new insights into the biological differences between the two tumor groups. Clin Cancer Res; 23(10); 2584-92. ©2016 AACR.

Harmonising and linking biomedical and clinical data across disparate data archives to enable integrative cross-biobank research.

A wealth of biospecimen samples are stored in modern globally distributed biobanks. Biomedical researchers worldwide need to be able to combine the available resources to improve the power of large-scale studies. A prerequisite for this effort is to be able to search and access phenotypic, clinical and other information about samples that are currently stored at biobanks in an integrated manner. However, privacy issues together with heterogeneous information systems and the lack of agreed-upon vocabularies have made specimen searching across multiple biobanks extremely challenging. We describe three case studies where we have linked samples and sample descriptions in order to facilitate global searching of available samples for research. The use cases include the ENGAGE (European Network for Genetic and Genomic Epidemiology) consortium comprising at least 39 cohorts, the SUMMIT (surrogate markers for micro- and macro-vascular hard endpoints for innovative diabetes tools) consortium and a pilot for data integration between a Swedish clinical health registry and a biobank. We used the Sample avAILability (SAIL) method for data linking: first, created harmonised variables and then annotated and made searchable information on the number of specimens available in individual biobanks for various phenotypic categories. By operating on this categorised availability data we sidestep many obstacles related to privacy that arise when handling real values and show that harmonised and annotated records about data availability across disparate biomedical archives provide a key methodological advance in pre-analysis exchange of information between biobanks, that is, during the project planning phase.

Interactions Between High- and Low-Risk HPV Types Reduce the Risk of Squamous Cervical Cancer.

BACKGROUND: The clinical significance of co-infections with high-risk (HR) and low-risk (LR) human papillomavirus (HPV) in the etiology of cervical cancer is debated, as prospective evidence on this issue is limited. However, the question is of increasing relevance in relation to HPV-based cancer prevention. METHODS: In two population-based nested case-control studies among women participating in cervical screening with baseline normal smears, we collected 4659 smears from women who later developed cancer in situ (CIS; n = 524) or squamous cervical cancer (SCC; n = 378) and individually matched control subjects who remained free of disease during study follow-up. The median follow-up until diagnosis was 6.4 to 7.8 years. All smears were tested for HPV. We used conditional logistic regression models with two-way interaction terms to estimate relative risks (RRs) for CIS and SCC, respectively. All statistical tests were two-sided. RESULTS: Compared with women who were infected with HRHPV only, women who were also infected with LRHPV had a lower risk for SCC (RR = 0.2, 95% confidence interval [CI] = 0.04 to 0.99, P = .049). This interaction was not shown for CIS (RR = 1.1, 95% CI = 0.4 to 3.6). Women who were positive for both HRHPV and LRHPV had, on average, a 4.8 year longer time to diagnosis of SCC than women who were positive for HRHPV only (P = .006). Results were highly robust in sensitivity analyses. CONCLUSION: Co-infection with LRHPV is associated with a lower risk of future invasive disease and longer time to diagnosis than infection with HRHPV alone. We propose that co-infection with LRHPV interferes with the rate of progression to invasive cervical cancer.

Case-control estimation of the impact of oncolytic adenovirus on the survival of patients with refractory solid tumors.

Oncolytic immunotherapy with cytokine armed replication competent viruses is an emerging approach in cancer treatment. In a recent randomized trial, an increase in response rate was seen but the effect on overall survival is not known with any virus. To facilitate randomized trials, we performed a case-control study assessing the survival of 270 patients treated in an Advanced Therapy Access Program (ATAP), in comparison to matched concurrent controls from the same hospital. The overall survival of all virus treated patients was not increased over controls. However, when analysis was restricted to GMCSF-sensitive tumor types treated with GMSCF-coding viruses, a significant improvement in median survival was present (from 170 to 208 days, P = 0.0012, N = 148). An even larger difference was seen when analysis was restricted to good performance score patients (193 versus 292 days, P = 0.034, N = 90). The survival of ovarian cancer patients was especially promising as median survival nearly quadrupled (P = 0.0003, N = 37). These preliminary data lend support to initiation of randomized clinical trials with GMCSF-coding oncolytic adenoviruses.

Radical prostatectomy or watchful waiting in early prostate cancer.

BACKGROUND: Radical prostatectomy reduces mortality among men with localized prostate cancer; however, important questions regarding long-term benefit remain. METHODS: Between 1989 and 1999, we randomly assigned 695 men with early prostate cancer to watchful waiting or radical prostatectomy and followed them through the end of 2012. The primary end points in the Scandinavian Prostate Cancer Group Study Number 4 (SPCG-4) were death from any cause, death from prostate cancer, and the risk of metastases. Secondary end points included the initiation of androgen-deprivation therapy. RESULTS: During 23.2 years of follow-up, 200 of 347 men in the surgery group and 247 of the 348 men in the watchful-waiting group died. Of the deaths, 63 in the surgery group and 99 in the watchful-waiting group were due to prostate cancer; the relative risk was 0.56 (95% confidence interval [CI], 0.41 to 0.77; P=0.001), and the absolute difference was 11.0 percentage points (95% CI, 4.5 to 17.5). The number needed to treat to prevent one death was 8. One man died after surgery in the radical-prostatectomy group. Androgen-deprivation therapy was used in fewer patients who underwent prostatectomy (a difference of 25.0 percentage points; 95% CI, 17.7 to 32.3). The benefit of surgery with respect to death from prostate cancer was largest in men younger than 65 years of age (relative risk, 0.45) and in those with intermediate-risk prostate cancer (relative risk, 0.38). However, radical prostatectomy was associated with a reduced risk of metastases among older men (relative risk, 0.68; P=0.04). CONCLUSIONS: Extended follow-up confirmed a substantial reduction in mortality after radical prostatectomy; the number needed to treat to prevent one death continued to decrease when the treatment was modified according to age at diagnosis and tumor risk. A large proportion of long-term survivors in the watchful-waiting group have not required any palliative treatment. (Funded by the Swedish Cancer Society and others.).

Prospective study of HPV16 viral load and risk of in situ and invasive squamous cervical cancer.

BACKGROUND: A strong association has been shown between high viral DNA load (VL) of human papillomavirus (HPV) type 16 and risk for cervical cancer in situ (CIS). However, little data is available for the significance of VL in invasive squamous cell carcinoma (SCC). METHODS: In 2 nested case-control studies among women participating in cervical screening, with a cytologically normal first smear, we collected 5,665 smears from 621 women with CIS, 457 with SCC, and individually matched controls. All smears were tested for HPV, and VLs of HPV16 positive smears were quantified using real time-PCR. The median follow-up until diagnosis of CIS or SCC was 6.1 to 7.7 years. RESULTS: Low VL's were common among both CIS and SCC case women, until 1 to 2 years before diagnosis when a surge in VL occurred. The relative risk (RR) associated with low viral load of HPV16 was around 10 for CIS, and 10 to 20 for SCC throughout 10 years before diagnosis, compared with HPV16-negative women. For women with medium to high VL, the risk for CIS was greatly increased from 5 years before diagnosis [RR, 19; 95% confidence interval (CI), 7-48]. In SCC, a high VL conferred an increased risk, but only from 3 years before diagnosis [RR, 60; 95% CI, 6-580]. CONCLUSIONS: We show differing risk functions associated with HPV16 viral load in CIS and SCC, respectively. We further show that viral loads were unexpectedly low early in the SCC disease process. IMPACT: HPV16 viral load appears highly complex which may limit its use in cervical screening.

Results from the Scandinavian Prostate Cancer Group Trial Number 4: a randomized controlled trial of radical prostatectomy versus watchful waiting.

In the Scandinavian Prostate Cancer Group Trial Number 4 (SPCG-4), 347 men were randomly assigned to radical prostatectomy and 348 to watchful waiting. In the most recent analysis (median follow-up time = 12.8 years), the cumulative mortality curves had been stable over the follow-up. At 15 years, the absolute risk reduction of dying from prostate cancer was 6.1% following randomization to radical prostatectomy, compared with watchful waiting. Hence, 17 need to be randomized to operation to avert one death. Data on self-reported symptoms, stress from symptoms, and quality of life were collected at 4 and 12.2 years of median follow-up. These questionnaire studies show an intricate pattern of symptoms evolving after surgery, hormonal treatments, signs of tumor progression, and also from natural aging. This article discusses some of the main findings of the SPCG-4 study.

Individualized estimation of the benefit of radical prostatectomy from the Scandinavian Prostate Cancer Group randomized trial.

BACKGROUND: Although there is randomized evidence that radical prostatectomy improves survival, there are few data on how benefit varies by baseline risk. OBJECTIVE: We aimed to create a statistical model to calculate the decrease in risk of death associated with surgery for an individual patient, using stage, grade, prostate-specific antigen, and age as predictors. DESIGN, SETTING, AND PARTICIPANTS: A total of 695 men with T1 or T2 prostate cancer participated in the Scandinavian Prostate Cancer Group 4 trial (SPCG-4). INTERVENTION: Patients in SPCG-4 were randomized to radical prostatectomy or conservative management. OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: Competing risk models were created separately for the radical prostatectomy and the watchful waiting group, with the difference between model predictions constituting the estimated benefit for an individual patient. RESULTS AND LIMITATIONS: Individualized predictions of surgery benefit varied widely depending on age and tumor characteristics. At 65 yr of age, the absolute 10-yr risk reduction in prostate cancer mortality attributable to radical prostatectomy ranged from 4.5% to 17.2% for low- versus high-risk patients. Little expected benefit was associated with surgery much beyond age 70. Only about a quarter of men had an individualized benefit within even 50% of the mean. A limitation is that estimates from SPCG-4 have to be applied cautiously to contemporary patients. CONCLUSIONS: Our model suggests that it is hard to justify surgery in patients with Gleason 6, T1 disease or in those patients much above 70 yr of age. Conversely, surgery seems unequivocally of benefit for patients who have Gleason 8, or Gleason 7, stage T2. For patients with Gleason 6 T2 and Gleason 7 T1, treatment is more of a judgment call, depending on patient preference and other clinical findings, such as the number of positive biopsy cores and comorbidities.

Prostate cancer risk variants are not associated with disease progression.

BACKGROUND: Currently used prognostic markers are limited in their ability to accurately predict disease progression among patients with localized prostate cancer. We examined 23 reported prostate cancer susceptibility variants for association with disease progression. METHODS: Disease progression was explored among 4,673 Swedish patients treated for clinically localized prostate cancer between 1997 and 2002. Prostate cancer progression was defined according to primary treatment as a composed event reflecting termination of deferred treatment, biochemical recurrence, local progression, or presence of distant metastasis. Association between single variants, and all variants combined, were performed in Cox regression analysis assuming both log-additive and co-dominant genetic models. RESULTS: Three of the 23 genetic variants explored were nominally associated with prostate cancer progression; rs9364554 (P = 0.041) on chromosome 6q25 and rs10896449 (P = 0.029) on chromosome 11q13 among patients treated with curative intent; and rs4054823 (P = 0.008) on chromosome 17p12 among patients on surveillance. However, none of these associations remained statistically significant after correction for multiple testing. The combined effect of all susceptibility variants was not associated with prostate cancer progression neither among patients receiving treatment with curative intent (P = 0.14) nor among patients on surveillance (P = 0.92). CONCLUSIONS: We observed no evidence for an association between any of 23 established prostate cancer genetic risk variants and disease progression. Accumulating evidence suggests separate genetic components for initiation and progression of prostate cancer. Future studies systematically searching for genetic risk variants associated with prostate cancer progression and prognosis are warranted.

The genetic structure of the Swedish population.

Patterns of genetic diversity have previously been shown to mirror geography on a global scale and within continents and individual countries. Using genome-wide SNP data on 5174 Swedes with extensive geographical coverage, we analyzed the genetic structure of the Swedish population. We observed strong differences between the far northern counties and the remaining counties. The population of Dalarna county, in north middle Sweden, which borders southern Norway, also appears to differ markedly from other counties, possibly due to this county having more individuals with remote Finnish or Norwegian ancestry than other counties. An analysis of genetic differentiation (based on pairwise F(st)) indicated that the population of Sweden's southernmost counties are genetically closer to the HapMap CEU samples of Northern European ancestry than to the populations of Sweden's northernmost counties. In a comparison of extended homozygous segments, we detected a clear divide between southern and northern Sweden with small differences between the southern counties and considerably more segments in northern Sweden. Both the increased degree of homozygosity in the north and the large genetic differences between the south and the north may have arisen due to a small population in the north and the vast geographical distances between towns and villages in the north, in contrast to the more densely settled southern parts of Sweden. Our findings have implications for future genome-wide association studies (GWAS) with respect to the matching of cases and controls and the need for within-county matching. We have shown that genetic differences within a single country may be substantial, even when viewed on a European scale. Thus, population stratification needs to be accounted for, even within a country like Sweden, which is often perceived to be relatively homogenous and a favourable resource for genetic mapping, otherwise inferences based on genetic data may lead to false conclusions.

Radical prostatectomy versus watchful waiting in early prostate cancer.

BACKGROUND: In 2008, we reported that radical prostatectomy, as compared with watchful waiting, reduces the rate of death from prostate cancer. After an additional 3 years of follow-up, we now report estimated 15-year results. METHODS: From October 1989 through February 1999, we randomly assigned 695 men with early prostate cancer to watchful waiting or radical prostatectomy. Follow-up was complete through December 2009, with histopathological review of biopsy and radical-prostatectomy specimens and blinded evaluation of causes of death. Relative risks, with 95% confidence intervals, were estimated with the use of a Cox proportional-hazards model. RESULTS: During a median of 12.8 years, 166 of the 347 men in the radical-prostatectomy group and 201 of the 348 in the watchful-waiting group died (P=0.007). In the case of 55 men assigned to surgery and 81 men assigned to watchful waiting, death was due to prostate cancer. This yielded a cumulative incidence of death from prostate cancer at 15 years of 14.6% and 20.7%, respectively (a difference of 6.1 percentage points; 95% confidence interval [CI], 0.2 to 12.0), and a relative risk with surgery of 0.62 (95% CI, 0.44 to 0.87; P=0.01). The survival benefit was similar before and after 9 years of follow-up, was observed also among men with low-risk prostate cancer, and was confined to men younger than 65 years of age. The number needed to treat to avert one death was 15 overall and 7 for men younger than 65 years of age. Among men who underwent radical prostatectomy, those with extracapsular tumor growth had a risk of death from prostate cancer that was 7 times that of men without extracapsular tumor growth (relative risk, 6.9; 95% CI, 2.6 to 18.4). CONCLUSIONS: Radical prostatectomy was associated with a reduction in the rate of death from prostate cancer. Men with extracapsular tumor growth may benefit from adjuvant local or systemic treatment. (Funded by the Swedish Cancer Society and the National Institutes of Health.).

LifeGene--a large prospective population-based study of global relevance.

Studying gene-environment interactions requires that the amount and quality of the lifestyle data is comparable to what is available for the corresponding genomic data. Sweden has several crucial prerequisites for comprehensive longitudinal biomedical research, such as the personal identity number, the universally available national health care system, continuously updated population and health registries and a scientifically motivated population. LifeGene builds on these strengths to bridge the gap between basic research and clinical applications with particular attention to populations, through a unique design in a research-friendly setting. LifeGene is designed both as a prospective cohort study and an infrastructure with repeated contacts of study participants approximately every 5 years. Index persons aged 18-45 years old will be recruited and invited to include their household members (partner and any children). A comprehensive questionnaire addressing cutting-edge research questions will be administered through the web with short follow-ups annually. Biosamples and physical measurements will also be collected at baseline, and re-administered every 5 years thereafter. Event-based sampling will be a key feature of LifeGene. The household-based design will give the opportunity to involve young couples prior to and during pregnancy, allowing for the first study of children born into cohort with complete pre-and perinatal data from both the mother and father. Questions and sampling schemes will be tailored to the participants' age and life events. The target of LifeGene is to enroll 500,000 Swedes and follow them longitudinally for at least 20 years.

Prospective study of human papillomavirus (HPV) types, HPV persistence, and risk of squamous cell carcinoma of the cervix.

BACKGROUND: The link between squamous cell cervical carcinoma and human papillomavirus (HPV) 16/18 is well established, but the magnitude of the risk association is uncertain and the importance of other high-risk HPV (HRHPV) types is unclear. METHODS: In two prospective nested case-control series among women participating in cytologic screening in Sweden, we collected 2,772 cervical smears from 515 women with cancer in situ (CIS), 315 with invasive squamous cell carcinoma (SCC), and individually matched controls. All smears were tested for HPV with PCR assays, and the median follow-up until diagnosis was 5 to 7 years. Conditional logistic regression was used to estimate relative risks (RR) and 95% confidence intervals (95% CI). RESULTS: The presence of HPV16/18 in the first smear was associated with 8.5-fold (95% CI, 5.3-13.7) and 18.6-fold (95% CI, 9.0-38.9) increased risks of CIS and SCC, respectively, compared with women negative for HPV. Infection with other HRHPV types in the first smear was also associated with significantly increased risks for both CIS and SCC. Persistence of HPV16 infection conferred a RR of 18.5 (95% CI, 6.5-52.9) for CIS and 19.5 (95% CI, 4.7-81.7) for SCC. The HPV16/18 attributable risk proportion was estimated at 30% to 50% for CIS, and 41% to 47% for SCC. Other HRHPV types also conferred significant proportions. CONCLUSIONS: Our large population-based study provides quantification of risks for different HPV types and prospective evidence that non-16/18 HRHPV types increase the risk for future cervical cancer. IMPACT: This study gives further insights into cervical cancer risk stratification with implications for HPV-based prevention strategies.

NordicDB: a Nordic pool and portal for genome-wide control data.

A cost-efficient way to increase power in a genetic association study is to pool controls from different sources. The genotyping effort can then be directed to large case series. The Nordic Control database, NordicDB, has been set up as a unique resource in the Nordic area and the data are available for authorized users through the web portal (http://www.nordicdb.org). The current version of NordicDB pools together high-density genome-wide SNP information from ∼5000 controls originating from Finnish, Swedish and Danish studies and shows country-specific allele frequencies for SNP markers. The genetic homogeneity of the samples was investigated using multidimensional scaling (MDS) analysis and pairwise allele frequency differences between the studies. The plot of the first two MDS components showed excellent resemblance to the geographical placement of the samples, with a clear NW-SE gradient. We advise researchers to assess the impact of population structure when incorporating NordicDB controls in association studies. This harmonized Nordic database presents a unique genome-wide resource for future genetic association studies in the Nordic countries.

Prospective study of human papillomavirus and risk of cervical adenocarcinoma.

Human papillomaviruses (HPV) are established as a major cause of cervical carcinoma. However, causality inference is dependent on prospective evidence showing that exposure predicts risk for future disease. Such evidence is available for squamous cell carcinoma, but not for cervical adenocarcinoma. We followed a population-based cohort of 994,120 women who participated in cytological screening in Sweden for a median of 6.7 years. Baseline smears from women who developed adenocarcinoma during follow-up (118 women with in situ disease and 164 with invasive disease) and their individually matched controls (1,434 smears) were analyzed for HPV using PCR. Conditional logistic regression was used to estimate odds ratios (OR) of future adenocarcinoma with 95% confidence intervals (CI). Being positive for HPV 16 in the first cytologically normal smear was associated with increased risks for both future adenocarcinoma in situ (OR: 11.0, 95% CI: 2.6-46.8) and invasive adenocarcinoma (OR: 16.0, 95% CI: 3.8-66.7), compared to being negative for HPV 16. Similarly, an HPV 18 positive smear was associated with increased risks for adenocarcinoma in situ (OR: 26.0, 95% CI: 3.5-192) and invasive adenocarcinoma (OR: 28.0, 95% CI: 3.8-206), compared to an HPV 18 negative smear. Being positive for HPV 16/18 in 2 subsequent smears was associated with an infinite risk of both in situ and invasive adenocarcinoma. In conclusion, infections with HPV 16 and 18 are detectable up to at least 14 years before diagnosis of cervical adenocarcinoma. Our data provide prospective evidence that the association of HPV 16/18 with cervical adenocarcinoma is strong and causal.

GENESTAT: an information portal for design and analysis of genetic association studies.

We present the rationale, the background and the structure for version 2.0 of the GENESTAT information portal (www.genestat.org) for statistical genetics. The fast methodological advances, coupled with a range of standalone software, makes it difficult for expert as well as non-expert users to orientate when designing and analysing their genetic studies. The ultimate ambition of GENESTAT is to guide on statistical methodology related to the broad spectrum of research in genetic epidemiology. GENESTAT 2.0 focuses on genetic association studies. Each entry provides a summary of a topic and gives links to key papers, websites and software. The flexibility of the internet is utilised for cross-referencing and for open editing. This paper gives an overview of GENESTAT and gives short introductions to the current main topics in GENESTAT, with additional entries on the website. Methods and software developers are invited to contribute to the portal, which is powered by a Wikipedia-type engine and allows easy additions and editing.

Sensitivity analysis for principal stratum direct effects, with an application to a study of physical activity and coronary heart disease.

SUMMARY: In many studies, the aim is to learn about the direct exposure effect, that is, the effect not mediated through an intermediate variable. For example, in circulation disease studies it may be of interest to assess whether a suitable level of physical activity can prevent disease, even if it fails to prevent obesity. It is well known that stratification on the intermediate may introduce a so-called posttreatment selection bias. To handle this problem, we use the framework of principal stratification (Frangakis and Rubin, 2002, Biometrics 58, 21-29) to define a causally relevant estimand--the principal stratum direct effect (PSDE). The PSDE is not identified in our setting. We propose a method of sensitivity analysis that yields a range of plausible values for the causal estimand. We compare our work to similar methods proposed in the literature for handling the related problem of "truncation by death."

A random change point model for assessing variability in repeated measures of cognitive function.

Some cognitive functions undergo transitions in old age, which motivates the use of a change point model for the individual trajectory. The age when the change occurs varies between individuals and is treated as random. We illustrate the properties of a random change point model and use it for data from a Swedish study of change in cognitive function in old age. Variance estimates are obtained from Markov chain Monte Carlo simulation using Gibbs sampling. The random change point model is compared with models within the family of linear random effects models. The focus is on the ability to capture variability in measures of cognitive function. The models make different assumptions about the variance over the age span, and we demonstrate that the random change point model has the most reasonable structure.